Pith. sign in

REVIEW 5 major objections 4 minor 84 references

DFYP: A Dynamic Fusion Framework with Spectral Channel Attention and Adaptive Operator learning for Crop Yield Prediction

T0 review · 5 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read DFYP, a dual-branch deep network that couples resolution-aware spectral channel attention with a learned operator-selection module for edge features and a learnable fusion of CNN and Vision Transformer outputs, claims consistent…

desk verdict A clear case of test-set tuning invalidates the headline performance claims; the AOL-Net description is also internally inconsistent, so the paper needs major revision before it is credible. read the letter →

arxiv 2507.05849 v1 pith:GPQNTA7O submitted 2025-07-08 cs.CV

Juli Zhang , Zeyu Yan , Jing Zhang , Qiguang Miao , Quan Wang This is my paper
classification cs.CV
keywords cropyieldpredictionremotesensingdynamicfusionspectralchannelattentionadaptiveoperatorlearningvisiontransformerconceptdriftmultispectralimagery
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper aims to show that a single architecture can outperform specialized baselines for crop yield prediction across different satellite resolutions, crop types, and years by making three parts of the model adaptive: spectral channel weighting, edge-operator choice, and the fusion of local and global features. On the coarse-resolution MODIS dataset it reports the best RMSE, MAE, and R² in six of seven test years, and on the high-resolution Sentinel-2 dataset it reports the best RMSE and MAE for all four crops, with the best R² except for cotton where MMST-ViT is marginally ahead. The central claim is that fixed handcrafted edge operators (Sobel, Scharr, etc.) and static fusion strategies leave performance on the table, and that learning to select operators and to weight branches dynamically yields more stable, transferable predictions. If correct, this would offer a practical, end-to-end trainable alternative to dataset-specific agricultural monitoring pipelines.

What carries the argument

The central object is the adaptive operator pool in AOL-Net: a library of Sobel, Scharr, and a learnable kernel initialized as a convex combination of the two, with a selection gate that picks exactly one operator per time step to compute an edge-enhanced map that is blended into the CNN input via a learnable scaling. RCA modifies the Squeeze-and-Excitation channel attention to use max pooling for low-resolution inputs and average pooling for high-resolution inputs, so that spectral reweighting depends on resolution. The third mechanism is the fusion function $f_\theta(X) = \alpha f_{\theta_1}(X) + \beta f_{\theta_2}(X)$ with $\alpha, \beta \in (0,1)$ learned by backpropagation; the paper deliberately relaxes the constraint $\alpha+\beta=1$ to let the model rescale branch contributions. All three are trained jointly by MSE backpropagation through the final fused prediction.

What would settle it

Inspect the operator-selection gate's outputs for every county in the test set. If the gate selects the same operator for every county within a given crop and year (i.e., the choice does not vary across inputs), then AOL-Net is effectively a validation-chosen fixed operator, and a fair comparison with the best fixed operator chosen on validation would reproduce the results. A model whose selected operator actually varies with input features would provide evidence for the claimed dynamic mechanism.

Watch

Extended reading notes

Core claim

DFYP's core discovery is that the combination of three adaptive mechanisms—resolution-aware channel attention, an operator-selection gate that picks among Sobel, Scharr, and a learnable kernel, and a loss-guided weighted fusion of a CNN branch and a Vision Transformer branch—produces consistently lower RMSE and MAE and higher R² than nine existing deep-learning baselines on two benchmark datasets. The paper argues that no single classical edge operator is best across years and crops, which it supports with experiments on eight fixed operators, and that a learnable fusion weight (not constrained to sum to one) stabilizes performance where concatenation or fixed averaging fluctuates. The claimed result is a framework that generalizes across coarse (MODIS) and fine (Sentinel-2) imagery and across soybean, corn, cotton, and winter wheat, with ablations showing each component contributes to the final gain.

Load-bearing premise

The load-bearing premise is that the adaptive operator selection is learned end-to-end by backpropagation, even though the paper specifies a hard, non-differentiable binary selection rule alongside a continuous backprop-trained interpolation and a validation-accuracy-based gate, and never reconciles these three formulations.

Editorial extensions

If this is right

  • A single DFYP-style architecture can replace dataset-specific handcrafted pipelines for county-level yield forecasting across different satellite sensors.
  • Because the model uses only pre-season and in-season observations, it is suitable for operational forecasts issued before harvest.
  • The dynamic fusion and operator selection reduce year-to-year error fluctuations, indicating a path toward models that tolerate concept drift in agricultural remote sensing.
  • The AOL ablation suggests that adaptive edge processing contributes more on high-resolution imagery, while ViT global modeling matters most on low-resolution imagery, guiding where to put architectural capacity.
  • If replicated, the reported gains would make DFYP a strong default baseline for future crop yield prediction research on the same datasets.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The unnormalized fusion weights ($\alpha$ and $\beta$ free in (0,1)) mean the model can rescale its own output; a fair ablation that normalizes them ($\alpha+\beta=1$) would reveal whether the reported robustness comes from the dynamic weighting or simply from extra scale flexibility.
  • The operator-selection gate in the paper is specified as hard binary selection but trained as a continuous interpolation in Eq. (3); a testable extension is a Gumbel-Softmax relaxation, which would make the whole selection genuinely differentiable and could be compared against the current validation-score-based gate.
  • The paper's empirical claim that fixed edge operators fluctuate with crop and year implies that other remote sensing regression tasks (e.g., biomass, evapotranspiration) may benefit from the same adaptive-operator plugin, a hypothesis that can be tested directly on those tasks.
  • Because the gate is described as selecting per time step, one could examine whether the chosen operator correlates with identifiable conditions (image date, phenological stage, drought index); if such correlations exist, the module could double as an interpretability tool for when edge emphasis matters.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

5 major / 4 minor

Summary. The manuscript proposes DFYP, a dual-branch crop yield prediction framework combining a Resolution-aware Channel Attention (RCA) module, an Adaptive Operator Learning Network (AOL-Net) for edge-operator selection, and a learnable fusion of CNN and ViT branches. Experiments on MODIS and Sentinel-2 report RMSE, MAE, and R² improvements over nine baselines. The authors claim consistent state-of-the-art performance across spatial resolutions, crop types, and years, attributing the gains to dynamic operator selection, resolution-aware spectral reweighting, and adaptive fusion.

Significance. Crop yield prediction from remote sensing is an important applied problem, and the paper addresses a real limitation of fixed edge operators in CNN-based models. The empirical comparison of eight handcrafted edge operators across years and crops is a useful motivation for adaptive operator selection. If the method were validated with a clean evaluation protocol, the dual-branch fusion and AOL ideas could be a practical engineering contribution. However, the manuscript provides no code, no random seeds, and no statistical tests, and the reported evidence is compromised by model selection on test years. As written, the significance of the claimed improvements cannot be assessed.

major comments (5)
  1. [Section IV.E, Table II] The hyperparameter analysis in Table II reports RMSE on the same years (2009–2015) that Section IV.A defines as the evaluation years for MODIS, and the final configuration (6 CNN layers, 8 attention heads, 4 ViT layers) is selected from this table. This means the architecture used in the headline comparison of Section IV.B was chosen using test labels, making the comparison circular. Any advantage of DFYP over fixed baselines could be an artifact of test-set tuning. The authors must select hyperparameters on a separate validation split (e.g., a subset of the training years) and then evaluate once on held-out years, or use nested cross-validation, and report results for the final model without any test-set feedback.
  2. [Section III.D, Eqs. (3)–(6)] The paper states in Eq. (3) that λ is optimized by backpropagation to interpolate between Sobel and Scharr, but Eq. (5) imposes hard selection with λ ∈ {0,1} and Eq. (4) selects the operator by validation accuracy scores. No mechanism is provided for gradients through the arg max or through the hard binary gate, so the joint end-to-end training claimed in Section III.F and Fig. 1 is not actually specified. If the operator is chosen by validation accuracy, AOL-Net is a model-selection rule rather than a learned component; if λ is learned continuously, Eq. (5) is violated. The authors need to reconcile these definitions, for example with a Gumbel-softmax or straight-through estimator and a clear description of how validation accuracy enters the training objective.
  3. [Section IV.B vs. Abstract] The abstract and conclusion claim DFYP “consistently outperforms” all baselines across all metrics, crops, and years, but the paper’s own results show that CNN+LSTM beats DFYP on all three metrics on MODIS in 2013 and that MMST-ViT has a higher R² for cotton on Sentinel-2. These exceptions are acknowledged in Section IV.B, so the global claim should be qualified to “best in most settings” or the evaluation should be restricted to the settings where the claim actually holds.
  4. [Section IV.A and Fig. 2] No variance or significance information is reported. All comparisons are single point estimates with no standard deviations, confidence intervals, or tests across random seeds. Given the small number of evaluation years (7 for MODIS) and crops (4 for Sentinel-2), the reported differences, especially the marginal cotton R² gap, cannot be distinguished from noise. The authors should report mean ± standard deviation over multiple runs and perform paired significance tests.
  5. [Section IV.A] For Sentinel-2, evaluation is performed on a single test year (2022), yet the paper claims robust performance “across time periods.” A one-year evaluation cannot support a temporal-robustness claim. The authors should either evaluate on multiple held-out years or remove the temporal-consistency claim for Sentinel-2.
minor comments (4)
  1. [Throughout] The manuscript contains numerous typos and grammatical errors, including “Yied” in the Section II.A heading, “repid development,” “comparision” in the Fig. 2 caption and Section IV.B, “abation” in Table III, and “DYFP” in Section III.B. A careful proofreading pass is needed.
  2. [Section III.D] The notation in Eqs. (4)–(6) is confusing: Eq. (4) defines K_t as an arg max over K, but Eq. (6) defines K_t as a convex combination with coefficients λ_Sobel, λ_Scharr, λ_Learnable. These two definitions are not obviously equivalent, and the relationship between the validation-based score S(K,t) and the coefficients should be made explicit.
  3. [Table II] Table II is difficult to read because the columns are years and the rows are parameter values without a clear indication of which configuration each number belongs to. Presenting the results as a table with one row per configuration and columns for RMSE per year (or aggregated RMSE) would greatly improve clarity.
  4. [Section III.C, Eq. (2)] The choice of max pooling for low-resolution images and average pooling for high-resolution images is stated without empirical justification or a reference. A short experiment or a more detailed rationale would help the reader understand why this resolution-dependent pooling is preferable.

Circularity Check

1 steps flagged · score 6.0 of 10

The MODIS hyperparameter configuration is selected using RMSE computed on the same 2009–2015 test years later used for the SOTA comparison, so part of the reported outperformance is selected on the benchmark itself.

  1. fitted input called prediction [Section IV.A (Experimental Setup) and Section IV.E (Hyperparameter Analysis), Table II]
    "To predict yield for a given yeart, the models are trained using data fromt start tot−1, with 90% of the multi-spectral remote sensing data used for training and 10% for validation. For MODIS-based experiments, models are trained using data fromt start = 2003 to t - 1, and evaluated on the years t ={2009–2015}. ... Table II illustrates the impact of these parameters on model performance, evaluated using RMSE ... For the MODIS dataset, the optimal configuration consists of 6 CNN layers, 8 attention heads, and 4 ViT layers."

    Section IV.A fixes 2009–2015 as the evaluation years for MODIS, and Section IV.E selects the final architecture (6 CNN layers, 8 attention heads, 4 ViT layers) as 'optimal' by inspecting RMSE values in Table II whose rows are exactly those same years. The DFYP numbers then reported against the baselines in Figure 2 are produced by the configuration that was chosen to minimize RMSE on the very years used for evaluation. This is the fitted-input-called-prediction pattern: the hyperparameters are fitted to the test labels, and the selected model's test-set performance is presented as an out-of-sample prediction.

full rationale

The paper contains no mathematical derivation that collapses to its inputs: RCA is a resolution-conditioned SE-style attention rule, AOL-Net is an operator-selection module (though its hard/soft selection equations are internally inconsistent), and the fusion is a learnable weighted sum with MSE loss. There are no load-bearing self-citations: the cited baselines and the Sentinel-2 dataset come from external groups, and no uniqueness theorem or prior-work ansatz is used to forbid alternatives. The one circularity-relevant step is in the evaluation protocol rather than the model equations: Table II is used to pick the MODIS hyperparameters from RMSE values computed on the same 2009–2015 years that are then used as test years in the SOTA comparison. That makes part of the headline empirical claim selected on the benchmark, i.e., a mild form of fitted input called prediction. The separate AOL-Net contradiction between backprop-trained continuous lambda and hard selection by validation accuracy is a correctness/implementation gap, not a circularity, because the operator gate uses validation data rather than the test labels.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The central empirical claim rests on standard supervised learning assumptions plus two specific premises: that the AOL operator gate can be trained end-to-end despite hard binary selection (Section III.D, Eqs. 3-6), and that hyperparameters were selected without using the test years (Section IV.E, Table II). The paper does not clearly satisfy either. There are five fitted scalar or kernel parameters (alpha, beta, gamma, lambda, K_learnable) whose final values are not reported. No new physical entities are introduced.

free parameters (5)
  • Fusion weight alpha (learned scalar in Eq. 12) = not reported
    Learned via backpropagation to weight the AOL-Net branch output against the ViT branch; alpha is a free parameter whose final value is not reported.
  • Fusion weight beta (learned scalar in Eq. 12) = not reported
    Learned via backpropagation to weight the ViT branch output; the paper relaxes alpha+beta=1, so both are independent free parameters.
  • Edge blend coefficient gamma (Eq. 7) = not reported
    Learnable scalar that mixes the original input with the edge-filtered map; its fitted value is not reported.
  • Operator interpolation lambda (Eq. 3) = not reported
    Presented as a backprop-trained continuous weight between Sobel and Scharr, but Eq. (5) later forces lambda to be binary; the paper never reports which mechanism was used or the fitted values.
  • Learnable edge kernel K_learnable = not reported
    A trainable convolution kernel initialized as a convex combination of Sobel and Scharr; its final values are not reported.
assumptions (4)
  • ad hoc to paper The operator selection gate in Eq. (4)-(5) is differentiable or has a surrogate gradient so the network can be trained end-to-end with backpropagation.
    Section III.D claims lambda is optimized via backpropagation, but hard binary selection (Eq. 5) is non-differentiable; no straight-through estimator is provided.
  • domain assumption The hyperparameters in Table I were selected on a validation split that is independent of the test years reported in the final comparison.
    Section IV.E evaluates hyperparameters using RMSE on years 2009-2015 (MODIS) and 2022 (Sentinel-2), which are the same years used for the final reported test performance. The paper does not state that a separate validation split was used.
  • domain assumption The validation data used to compute the historical performance score S(K,t) in Eq. (4) do not include the test years.
    The paper does not specify the temporal split for the operator selection; if S(K,t) incorporates test-year accuracy, the selected operator is fitted to the test set.
  • standard math Standard supervised learning assumptions: the MODIS and Sentinel-2 datasets are representative and the USDA county yield labels are accurate.
    The empirical evaluation treats the datasets and labels as ground truth without error analysis.

how reviews work

0 comments
Cite this review

Pith. "Pith review of DFYP: A Dynamic Fusion Framework with Spectral Channel Attention and Adaptive Operator learning for Crop Yield Prediction." pith.science (2026). https://pith.science/paper/GPQNTA7O

@misc{pith2026250705849,
  author       = {Pith},
  title        = {Pith review of: DFYP: A Dynamic Fusion Framework with Spectral Channel Attention and Adaptive Operator learning for Crop Yield Prediction},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GPQNTA7O}},
  note         = {Machine review of arXiv:2507.05849}
}
read the original abstract

Accurate remote sensing-based crop yield prediction remains a fundamental challenging task due to complex spatial patterns, heterogeneous spectral characteristics, and dynamic agricultural conditions. Existing methods often suffer from limited spatial modeling capacity, weak generalization across crop types and years. To address these challenges, we propose DFYP, a novel Dynamic Fusion framework for crop Yield Prediction, which combines spectral channel attention, edge-adaptive spatial modeling and a learnable fusion mechanism to improve robustness across diverse agricultural scenarios. Specifically, DFYP introduces three key components: (1) a Resolution-aware Channel Attention (RCA) module that enhances spectral representation by adaptively reweighting input channels based on resolution-specific characteristics; (2) an Adaptive Operator Learning Network (AOL-Net) that dynamically selects operators for convolutional kernels to improve edge-sensitive spatial feature extraction under varying crop and temporal conditions; and (3) a dual-branch architecture with a learnable fusion mechanism, which jointly models local spatial details and global contextual information to support cross-resolution and cross-crop generalization. Extensive experiments on multi-year datasets MODIS and multi-crop dataset Sentinel-2 demonstrate that DFYP consistently outperforms current state-of-the-art baselines in RMSE, MAE, and R2 across different spatial resolutions, crop types, and time periods, showcasing its effectiveness and robustness for real-world agricultural monitoring.

Figures

Figures reproduced from arXiv: 2507.05849 by the authors.

Figure 1
Figure 1. Overview of the proposed DFYP framework for remote sensing-based crop yield prediction. The input consists of multi-spectral remote sensing [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Comparision with state-of-the-art methods. Fig. (a, b, c) show the RMSE, MAE and [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Performance of operators on MODIS and Sentinel-2. Comparison [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: Performance of Different Network Architectures [PITH_FULL_IMAGE:figures/full_fig_p009_4.png]
Figure 5
Figure 5. Figure 5: Ablation study results of different components on the MODIS and Sentinel-2 dataset. [PITH_FULL_IMAGE:figures/full_fig_p010_5.png]
Figure 6
Figure 6. Figure 6: Spatiotemporal Error Maps for Comparative Model Analysis [PITH_FULL_IMAGE:figures/full_fig_p012_6.png]
Figure 7
Figure 7. Figure 7: Visualization of the Sentinel-2 dataset. [PITH_FULL_IMAGE:figures/full_fig_p013_7.png]
Figure 8
Figure 8. Figure 8: Example image of Sentinel-2 data with different time. [PITH_FULL_IMAGE:figures/full_fig_p013_8.png]
Figure 9
Figure 9. Figure 9: Histogrammed image of MODIS data. The upper row shows the [PITH_FULL_IMAGE:figures/full_fig_p013_9.png]
Figure 8
Figure 8. Figure 8: In this supplementary material, we provide a comprehensive foundation that supports the main paper’s findings, offering deeper insights into the data characteristics and processing methodologies of our DFYP model for crop yield prediction from remote sensing imagery. R…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

84 extracted references · 60 canonical work pages

  1. [1]

    Machine learning regression techniques for the silage maize yield prediction using time-series images of landsat 8 oli,

    H. Aghighi, M. Azadbakht, D. Ashourloo, H. S. Shahrabi, and S. Ra- diom, “Machine learning regression techniques for the silage maize yield prediction using time-series images of landsat 8 oli,”IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 11, no. 12, pp. 4563–4577, 2018

    2018

  2. [2]

    Forecasting yield by integrating agrarian factors and machine learning models: A survey,

    D. Elavarasan, D. R. Vincent, V . Sharma, A. Y . Zomaya, and K. Srini- vasan, “Forecasting yield by integrating agrarian factors and machine learning models: A survey,”Computers and electronics in agriculture, vol. 155, pp. 257–282, 2018

    2018

  3. [3]

    Kstage: A knowledge-guided spatial-temporal attention graph learning network for crop yield prediction,

    M. Qiao, X. He, X. Cheng, P. Li, Q. Zhao, C. Zhao, and Z. Tian, “Kstage: A knowledge-guided spatial-temporal attention graph learning network for crop yield prediction,”Information Sciences, vol. 619, pp. 19–37, 2023

    2023

  4. [4]

    Crop yield estimation using time-series modis data and the effects of cropland masks in ontario, canada,

    J. Liu, J. Shang, B. Qian, T. Huffman, Y . Zhang, T. Dong, Q. Jing, and T. Martin, “Crop yield estimation using time-series modis data and the effects of cropland masks in ontario, canada,”Remote Sensing, vol. 11, no. 20, p. 2419, 2019

    2019

  5. [5]

    Crop yield prediction in agriculture: A comprehensive review of machine learning and deep learning ap- proaches, with insights for future research and sustainability,

    M. A. Jabed and M. A. A. Murad, “Crop yield prediction in agriculture: A comprehensive review of machine learning and deep learning ap- proaches, with insights for future research and sustainability,”Heliyon, 2024

    work page 2024

  6. [6]

    Corn yield prediction with ensemble cnn-dnn,

    M. Shahhosseini, G. Hu, S. Khaki, and S. V . Archontoulis, “Corn yield prediction with ensemble cnn-dnn,”Frontiers in plant science, vol. 12, p. 709008, 2021

    work page 2021

  7. [7]

    Crop yield prediction: An operational approach to crop yield modeling on field and subfield level with machine learning models,

    P. Helber, B. Bischke, P. Habelitz, C. Sanchez, D. Pathak, M. Miranda, H. Najjar, F. Mena, J. Siddamsetty, D. Arenaset al., “Crop yield prediction: An operational approach to crop yield modeling on field and subfield level with machine learning models,” inIGARSS 2023- 2023 IEEE International Geoscience and Remote Sensing Symposium. IEEE, 2023, pp. 2763–2766

    work page 2023

  8. [8]

    Deep gaussian process for crop yield prediction based on remote sensing data,

    J. You, X. Li, M. Low, D. Lobell, and S. Ermon, “Deep gaussian process for crop yield prediction based on remote sensing data,” inProceedings of the AAAI conference on artificial intelligence, vol. 31, no. 1, 2017

    work page 2017

Show all 84 references
  1. [9]

    A cnn-rnn framework for crop yield prediction,

    S. Khaki, L. Wang, and S. V . Archontoulis, “A cnn-rnn framework for crop yield prediction,”Frontiers in Plant Science, vol. 10, p. 1750, 2020

    2020

  2. [10]

    Deep learning for remote sensing data: A technical tutorial on the state of the art,

    L. Zhang, L. Zhang, and B. Du, “Deep learning for remote sensing data: A technical tutorial on the state of the art,”IEEE Geoscience and remote sensing magazine, vol. 4, no. 2, pp. 22–40, 2016

    2016

  3. [11]

    Transformers in remote sensing: A survey,

    A. A. Aleissaee, A. Kumar, R. M. Anwer, S. Khan, H. Cholakkal, G.- S. Xia, and F. S. Khan, “Transformers in remote sensing: A survey,” Remote Sensing, vol. 15, no. 7, p. 1860, 2023

    2023

  4. [12]

    An image is worth 16x16 words: Transformers for image recognition at scale,

    A. Dosovitskiy, “An image is worth 16x16 words: Transformers for image recognition at scale,”arXiv preprint arXiv:2010.11929, 2020

    2010 arXiv

  5. [13]

    Deep learning in remote sensing applications: A meta-analysis and review,

    L. Ma, Y . Liu, X. Zhang, Y . Ye, G. Yin, and B. A. Johnson, “Deep learning in remote sensing applications: A meta-analysis and review,” ISPRS journal of photogrammetry and remote sensing, vol. 152, pp. 166–177, 2019

    2019

  6. [14]

    Levit: a vision transformer in convnet’s clothing for faster inference,

    B. Graham, A. El-Nouby, H. Touvron, P. Stock, A. Joulin, H. J ´egou, and M. Douze, “Levit: a vision transformer in convnet’s clothing for faster inference,” inProceedings of the IEEE/CVF international conference on computer vision, 2021, pp. 12 259–12 269

    2021

  7. [15]

    Crop yield prediction using deep neural networks,

    S. Khaki and L. Wang, “Crop yield prediction using deep neural networks,”Frontiers in plant science, vol. 10, p. 621, 2019

    2019

  8. [16]

    Satellite-based soybean yield forecast: Integrating machine learning and weather data for improving crop yield prediction in southern brazil,

    R. A. Schwalbert, T. Amado, G. Corassa, L. P. Pott, P. V . Prasad, and I. A. Ciampitti, “Satellite-based soybean yield forecast: Integrating machine learning and weather data for improving crop yield prediction in southern brazil,”Agricultural and Forest Meteorology, vol. 284,...

    2020

  9. [17]

    Deep learning for multi-source data-driven crop yield prediction in northeast china,

    J. Lu, J. Li, H. Fu, X. Tang, Z. Liu, H. Chen, Y . Sun, and X. Ning, “Deep learning for multi-source data-driven crop yield prediction in northeast china,”Agriculture, vol. 14, no. 6, p. 794, 2024

    2024

  10. [18]

    Edge-guided recurrent positioning network for salient object detection in optical remote sensing images,

    X. Zhou, K. Shen, L. Weng, R. Cong, B. Zheng, J. Zhang, and C. Yan, “Edge-guided recurrent positioning network for salient object detection in optical remote sensing images,”IEEE Transactions on Cybernetics, vol. 53, no. 1, pp. 539–552, 2022

    2022

  11. [19]

    Revisiting edge detection in convolutional neural networks,

    M. Le and S. Kayal, “Revisiting edge detection in convolutional neural networks,” in2021 International Joint Conference on Neural Networks (IJCNN). IEEE, 2021, pp. 1–9

    2021

  12. [20]

    Learning under concept drift: A review,

    J. Lu, A. Liu, F. Dong, F. Gu, J. Gama, and G. Zhang, “Learning under concept drift: A review,”IEEE transactions on knowledge and data engineering, vol. 31, no. 12, pp. 2346–2363, 2018. JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2021 15

    2018

  13. [21]

    A survey on concept drift adaptation,

    J. Gama, I. ˇZliobait˙e, A. Bifet, M. Pechenizkiy, and A. Bouchachia, “A survey on concept drift adaptation,”ACM computing surveys (CSUR), vol. 46, no. 4, pp. 1–37, 2014

    2014

  14. [22]

    Deep learning in agriculture: A survey,

    A. Kamilaris and F. X. Prenafeta-Bold ´u, “Deep learning in agriculture: A survey,”Computers and electronics in agriculture, vol. 147, pp. 70–90, 2018

    2018

  15. [23]

    Deep learning for crop yield prediction: a systematic literature review,

    A. Oikonomidis, C. Catal, and A. Kassahun, “Deep learning for crop yield prediction: a systematic literature review,”New Zealand Journal of Crop and Horticultural Science, vol. 51, no. 1, pp. 1–26, 2023

    2023

  16. [24]

    Crop yield predic- tion using machine learning and deep learning techniques,

    K. Jhajharia, P. Mathur, S. Jain, and S. Nijhawan, “Crop yield predic- tion using machine learning and deep learning techniques,”Procedia Computer Science, vol. 218, pp. 406–417, 2023

    2023

  17. [25]

    Deep learning,

    Y . LeCun, Y . Bengio, and G. Hinton, “Deep learning,”nature, vol. 521, no. 7553, pp. 436–444, 2015

    2015

  18. [26]

    You only look once: Unified, real-time object detection,

    J. Redmon, “You only look once: Unified, real-time object detection,” inProceedings of the IEEE conference on computer vision and pattern recognition, 2016

    2016

  19. [27]

    Simultaneous corn and soybean yield prediction from remote sensing data using deep transfer learning,

    S. Khaki, H. Pham, and L. Wang, “Simultaneous corn and soybean yield prediction from remote sensing data using deep transfer learning,” Scientific Reports, vol. 11, no. 1, p. 11132, 2021

    2021

  20. [28]

    Deep learning for vegetation health forecasting: a case study in kenya,

    T. Lees, G. Tseng, C. Atzberger, S. Reece, and S. Dadson, “Deep learning for vegetation health forecasting: a case study in kenya,”Remote Sensing, vol. 14, no. 3, p. 698, 2022

    2022

  21. [29]

    Panoptic segmentation of satellite image time series with convolutional temporal attention networks,

    V . S. F. Garnot and L. Landrieu, “Panoptic segmentation of satellite image time series with convolutional temporal attention networks,” in Proceedings of the IEEE/CVF International Conference on Computer Vision, 2021, pp. 4872–4881

    2021

  22. [30]

    A gnn-rnn approach for harnessing geospatial and temporal information: applica- tion to crop yield prediction,

    J. Fan, J. Bai, Z. Li, A. Ortiz-Bobea, and C. P. Gomes, “A gnn-rnn approach for harnessing geospatial and temporal information: applica- tion to crop yield prediction,” inProceedings of the AAAI conference on artificial intelligence, vol. 36, no. 11, 2022, pp. 11 873–11 881

    2022

  23. [31]

    A time-dependent parameter estimation framework for crop modeling,

    F. Akhavizadegan, J. Ansarifar, L. Wang, I. Huber, and S. V . Archon- toulis, “A time-dependent parameter estimation framework for crop modeling,”Scientific reports, vol. 11, no. 1, p. 11437, 2021

    2021

  24. [32]

    Advanc- ing agricultural research using machine learning algorithms,

    S. Mourtzinis, P. D. Esker, J. E. Specht, and S. P. Conley, “Advanc- ing agricultural research using machine learning algorithms,”Scientific reports, vol. 11, no. 1, p. 17879, 2021

    2021

  25. [33]

    Learning long- term crop management strategies with cyclesgym,

    M. Turchetta, L. Corinzia, S. Sussex, A. Burton, J. Herrera, I. Athanasiadis, J. M. Buhmann, and A. Krause, “Learning long- term crop management strategies with cyclesgym,”Advances in neural information processing systems, vol. 35, pp. 11 396–11 409, 2022

    2022

  26. [34]

    Crop yield prediction with deep convolutional neural networks,

    P. Nevavuori, N. Narra, and T. Lipping, “Crop yield prediction with deep convolutional neural networks,”Computers and electronics in agriculture, vol. 163, p. 104859, 2019

    2019

  27. [35]

    Deep learning-based classification of hyperspectral data,

    Y . Chen, Z. Lin, X. Zhao, G. Wang, and Y . Gu, “Deep learning-based classification of hyperspectral data,”IEEE Journal of Selected topics in applied earth observations and remote sensing, vol. 7, no. 6, pp. 2094–2107, 2014

    2014

  28. [36]

    Enhancing rice crop health: a light weighted cnn-based disease detection system with mobile application integration,

    S. Tyagi, S. Reddy, R. Anand, and A. Sabharwal, “Enhancing rice crop health: a light weighted cnn-based disease detection system with mobile application integration,”Multimedia Tools and Applications, vol. 83, no. 16, pp. 48 799–48 829, 2024

    2024

  29. [37]

    Comparison of cnn models for application in crop health assessment with participatory sensing,

    P. Bhatt, S. Sarangi, and S. Pappula, “Comparison of cnn models for application in crop health assessment with participatory sensing,” in2017 IEEE Global Humanitarian Technology Conference (GHTC). IEEE, 2017, pp. 1–7

    2017

  30. [38]

    Exploring the synergies of hybrid cnns and vits architectures for computer vision: A survey,

    H. Yunusa, S. Qin, A. H. A. Chukkol, A. A. Yusuf, I. Bello, and A. Lawan, “Exploring the synergies of hybrid cnns and vits architectures for computer vision: A survey,”arXiv preprint arXiv:2402.02941, 2024

    2024

  31. [39]

    Restormer: Efficient transformer for high-resolution image restoration,

    S. W. Zamir, A. Arora, S. Khan, M. Hayat, F. S. Khan, and M.-H. Yang, “Restormer: Efficient transformer for high-resolution image restoration,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2022, pp. 5728–5739

    2022

  32. [40]

    Attention is all you need,

    A. Vaswani, “Attention is all you need,”Advances in Neural Information Processing Systems, 2017

    2017

  33. [41]

    Self-attention generative adversarial networks,

    H. Zhang, I. Goodfellow, D. Metaxas, and A. Odena, “Self-attention generative adversarial networks,” inInternational conference on machine learning. PMLR, 2019, pp. 7354–7363

    2019

  34. [42]

    Transformers in vision: A survey,

    S. Khan, M. Naseer, M. Hayat, S. W. Zamir, F. S. Khan, and M. Shah, “Transformers in vision: A survey,”ACM computing surveys (CSUR), vol. 54, no. 10s, pp. 1–41, 2022

    2022

  35. [43]

    Extended vision transformer (exvit) for land use and land cover classification: A mul- timodal deep learning framework,

    J. Yao, B. Zhang, C. Li, D. Hong, and J. Chanussot, “Extended vision transformer (exvit) for land use and land cover classification: A mul- timodal deep learning framework,”IEEE Transactions on Geoscience and Remote Sensing, vol. 61, pp. 1–15, 2023

    2023

  36. [44]

    Mmst-vit: Climate change- aware crop yield prediction via multi-modal spatial-temporal vision transformer,

    F. Lin, S. Crawford, K. Guillot, Y . Zhang, Y . Chen, X. Yuan, L. Chen, S. Williams, R. Minvielle, X. Xiaoet al., “Mmst-vit: Climate change- aware crop yield prediction via multi-modal spatial-temporal vision transformer,” inProceedings of the IEEE/CVF International Conference...

    2023

  37. [45]

    Enhancing crop productivity and sustainability through dis- ease identification in maize leaves: Exploiting a large dataset with an advanced vision transformer model,

    I. Pacal, “Enhancing crop productivity and sustainability through dis- ease identification in maize leaves: Exploiting a large dataset with an advanced vision transformer model,”Expert Systems with Applications, vol. 238, p. 122099, 2024

    2024

  38. [46]

    Conformer: Local features coupling global representations for visual recognition,

    Z. Peng, W. Huang, S. Gu, L. Xie, Y . Wang, J. Jiao, and Q. Ye, “Conformer: Local features coupling global representations for visual recognition,” inProceedings of the IEEE/CVF international conference on computer vision, 2021, pp. 367–376

    2021

  39. [47]

    Deepyield: A com- bined convolutional neural network with long short-term memory for crop yield forecasting,

    K. Gavahi, P. Abbaszadeh, and H. Moradkhani, “Deepyield: A com- bined convolutional neural network with long short-term memory for crop yield forecasting,”Expert Systems with Applications, vol. 184, p. 115511, Dec. 2021

    2021

  40. [48]

    Cvt: Introducing convolutions to vision transformers,

    H. Wu, B. Xiao, N. Codella, M. Liu, X. Dai, L. Yuan, and L. Zhang, “Cvt: Introducing convolutions to vision transformers,” inProceedings of the IEEE/CVF international conference on computer vision, 2021, pp. 22–31

    2021

  41. [49]

    Incorporating convolution designs into visual transformers,

    K. Yuan, S. Guo, Z. Liu, A. Zhou, F. Yu, and W. Wu, “Incorporating convolution designs into visual transformers,” inProceedings of the IEEE/CVF international conference on computer vision, 2021, pp. 579– 588

    2021

  42. [50]

    Escaping the big data paradigm with compact transformers,

    A. Hassani, S. Walton, N. Shah, A. Abuduweili, J. Li, and H. Shi, “Escaping the big data paradigm with compact transformers,”arXiv preprint arXiv:2104.05704, 2021

    2021 arXiv

  43. [51]

    Localvit: Bring- ing locality to vision transformers,

    Y . Li, K. Zhang, J. Cao, R. Timofte, and L. Van Gool, “Localvit: Bring- ing locality to vision transformers,”arXiv preprint arXiv:2104.05707, 2021

    2021 arXiv

  44. [52]

    Aggregating nested transformers,

    Z. Zhang, H. Zhang, L. Zhao, T. Chen, and T. Pfister, “Aggregating nested transformers,”arXiv preprint arXiv:2105.12723, vol. 2, no. 3, p. 5, 2021

    2021 arXiv

  45. [53]

    End-to-end object detection with transformers,

    N. Carion, F. Massa, G. Synnaeve, N. Usunier, A. Kirillov, and S. Zagoruyko, “End-to-end object detection with transformers,” in European conference on computer vision. Springer, 2020, pp. 213– 229

    2020

  46. [54]

    Mobile-former: Bridging mobilenet and transformer,

    Y . Chen, X. Dai, D. Chen, M. Liu, X. Dong, L. Yuan, and Z. Liu, “Mobile-former: Bridging mobilenet and transformer,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2022, pp. 5270–5279

    2022

  47. [55]

    Tccnet: Temporally consistent context-free network for semi- supervised video polyp segmentation

    X. Li, J. Xu, Y . Zhang, R. Feng, R.-W. Zhao, T. Zhang, X. Lu, and S. Gao, “Tccnet: Temporally consistent context-free network for semi- supervised video polyp segmentation.” inIJCAI, 2022, pp. 1109–1115

    2022

  48. [56]

    Cross- enhancement transformer for action segmentation,

    J. Wang, Z. Wang, S. Zhuang, Y . Hao, and H. Wang, “Cross- enhancement transformer for action segmentation,”Multimedia Tools and Applications, vol. 83, no. 9, pp. 25 643–25 656, 2024

    2024

  49. [57]

    Coatnet: Marrying convolu- tion and attention for all data sizes,

    Z. Dai, H. Liu, Q. V . Le, and M. Tan, “Coatnet: Marrying convolu- tion and attention for all data sizes,”Advances in neural information processing systems, vol. 34, pp. 3965–3977, 2021

    2021

  50. [58]

    Cmt: Convolutional neural networks meet vision transformers. 2022 ieee,

    J. Guo, K. Han, H. Wu, C. Xu, Y . Tang, C. Xu, and Y . Wang, “Cmt: Convolutional neural networks meet vision transformers. 2022 ieee,” in CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022, pp. 12 165–12 175

    2022

  51. [59]

    Vitae: Vision transformer advanced by exploring intrinsic inductive bias,

    Y . Xu, Q. Zhang, J. Zhang, and D. Tao, “Vitae: Vision transformer advanced by exploring intrinsic inductive bias,”Advances in neural information processing systems, vol. 34, pp. 28 522–28 535, 2021

    2021

  52. [60]

    Dilated neighborhood attention transformer,

    A. Hassani and H. Shi, “Dilated neighborhood attention transformer,” arXiv preprint arXiv:2209.15001, 2022

    2022 arXiv

  53. [61]

    Per-pixel classification is not all you need for semantic segmentation,

    B. Cheng, A. Schwing, and A. Kirillov, “Per-pixel classification is not all you need for semantic segmentation,”Advances in neural information processing systems, vol. 34, pp. 17 864–17 875, 2021

    2021

  54. [62]

    Efficientformer: Vision transformers at mobilenet speed,

    Y . Li, G. Yuan, Y . Wen, J. Hu, G. Evangelidis, S. Tulyakov, Y . Wang, and J. Ren, “Efficientformer: Vision transformers at mobilenet speed,”Ad- vances in Neural Information Processing Systems, vol. 35, pp. 12 934– 12 949, 2022

    2022

  55. [63]

    Swin2sr: Swinv2 transformer for compressed image super-resolution and restoration,

    M. V . Conde, U.-J. Choi, M. Burchi, and R. Timofte, “Swin2sr: Swinv2 transformer for compressed image super-resolution and restoration,” in European Conference on Computer Vision. Springer, 2022, pp. 669– 687

    2022

  56. [64]

    Vitmatte: Boosting image matting with pre-trained plain vision transformers,

    J. Yao, X. Wang, S. Yang, and B. Wang, “Vitmatte: Boosting image matting with pre-trained plain vision transformers,”Information Fusion, vol. 103, p. 102091, 2024

    2024

  57. [65]

    Batalik, F

    L. Batalik, F. Dosbaba, M. Hartman, V . Konecny, K. Batalikova, and J. Spinar, “Long-term exercise effects after cardiac telerehabilitation in patients with coronary artery disease: 1-year follow-up results of the JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2021 16 ra...

    2021

  58. [66]

    Eca-net: Efficient channel attention for deep convolutional neural networks,

    Q. Wang, B. Wu, P. Zhu, P. Li, W. Zuo, and Q. Hu, “Eca-net: Efficient channel attention for deep convolutional neural networks,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2020, pp. 11 534–11 542

    2020

  59. [67]

    Dual attention network for scene segmentation,

    J. Fu, J. Liu, H. Tian, Y . Li, Y . Bao, Z. Fang, and H. Lu, “Dual attention network for scene segmentation,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2019, pp. 3146– 3154

    2019

  60. [68]

    Concurrent spatial and channel ‘squeeze & excitation’in fully convolutional networks,

    A. G. Roy, N. Navab, and C. Wachinger, “Concurrent spatial and channel ‘squeeze & excitation’in fully convolutional networks,” inMedical Image Computing and Computer Assisted Intervention–MICCAI 2018: 21st International Conference, Granada, Spain, September 16-20, 2018, Procee...

    2018

  61. [69]

    Cbam: Convolutional block attention module,

    S. Woo, J. Park, J.-Y . Lee, and I. S. Kweon, “Cbam: Convolutional block attention module,” inProceedings of the European conference on computer vision (ECCV), 2018, pp. 3–19

    2018

  62. [70]

    Bottleneck transformers for visual recognition,

    A. Srinivas, T.-Y . Lin, N. Parmar, J. Shlens, P. Abbeel, and A. Vaswani, “Bottleneck transformers for visual recognition,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2021, pp. 16 519–16 529

    2021

  63. [71]

    Crop yield prediction using multitemporal uav data and spatio-temporal deep learning models,

    P. Nevavuori, N. Narra, P. Linna, and T. Lipping, “Crop yield prediction using multitemporal uav data and spatio-temporal deep learning models,” Remote Sensing, vol. 12, no. 23, p. 4000, Dec. 2020

    2020

  64. [72]

    Multispectral crop yield prediction using 3dconvolutional neural networks and attention convolutional lstm approaches,

    S. M. M. Nejad, D. Abbasi-Moghadam, A. Sharifi, N. Farmonov, K. Amankulova, and M. Laszlz, “Multispectral crop yield prediction using 3dconvolutional neural networks and attention convolutional lstm approaches,”IEEE Journal of Selected Topics in Applied Earth Obser- vations an...

    2023

  65. [73]

    Con- vlstm–vit: A deep neural network for crop yield prediction using earth observations and remotely sensed data,

    S. M. Mirhoseini Nejad, D. Abbasi-Moghadam, and A. Sharifi, “Con- vlstm–vit: A deep neural network for crop yield prediction using earth observations and remotely sensed data,”IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 17, pp. 17 489...

    2024

  66. [74]

    Squeeze-and-excitation networks,

    J. Hu, L. Shen, and G. Sun, “Squeeze-and-excitation networks,” in Proceedings of the IEEE conference on computer vision and pattern recognition, 2018, pp. 7132–7141

    2018

  67. [75]

    An open and large-scale dataset for multi-modal climate change-aware crop yield predictions,

    F. Lin, K. Guillot, S. Crawford, Y . Zhang, X. Yuan, and N.-F. Tzeng, “An open and large-scale dataset for multi-modal climate change-aware crop yield predictions,” inProceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, 2024, pp. 5375–5386

    2024

  68. [76]

    A convlstm conjunction model for groundwater level forecasting in a karst aquifer considering connectivity characteristics,

    F. Guo, J. Yang, H. Li, G. Li, and Z. Zhang, “A convlstm conjunction model for groundwater level forecasting in a karst aquifer considering connectivity characteristics,”Water, vol. 13, no. 19, p. 2759, Oct. 2021

    2021

  69. [77]

    Agricultural trade publications and the 2012 midwestern us drought: A missed opportunity for climate risk communication,

    S. P. Church, T. Haigh, M. Widhalm, S. G. De Jalon, N. Babin, J. S. Carlton, M. Dunn, K. Fagan, C. L. Knutson, and L. S. Prokopy, “Agricultural trade publications and the 2012 midwestern us drought: A missed opportunity for climate risk communication,”Climate Risk Management, ...

    2012

  70. [78]

    A 3x3 isotropic gradient operator for image processing,

    I. Sobel, G. Feldmanet al., “A 3x3 isotropic gradient operator for image processing,”a talk at the Stanford Artificial Project in, vol. 1968, pp. 271–272, 1968

    1968

  71. [79]

    A computational approach to edge detection,

    J. Canny, “A computational approach to edge detection,”IEEE Transac- tions on Pattern Analysis and Machine Intelligence, vol. PAMI-8, no. 6, pp. 679–698, 1986

    1986

  72. [80]

    Computer determination of the constituent structure of biological images,

    R. A. Kirsch, “Computer determination of the constituent structure of biological images,” inComputers and Biomedical Research, vol. 4, no. 3. Elsevier, 1971, pp. 315–328

    1971

  73. [81]

    Theory of edge detection,

    D. Marr and E. Hildreth, “Theory of edge detection,”Proceedings of the Royal Society of London. Series B. Biological Sciences, vol. 207, no. 1167, pp. 187–217, 1980

    1980

  74. [82]

    Object enhancement and extraction,

    J. M. Prewitt, “Object enhancement and extraction,”Picture Processing and Psychopictorics, vol. 10, no. 1, pp. 15–19, 1970

    1970

  75. [83]

    Machine perception of three-dimensional, so lids,

    P. E. DO CT OR OF, “Machine perception of three-dimensional, so lids,” Ph.D. dissertation, PhD thesis, MASSACHUSETTS INSTITUTE OF TECHNOLOGY , 1961

    1961

  76. [84]

    Optimal filters for extended optical flow,

    H. Scharr, “Optimal filters for extended optical flow,” inInternational Workshop on Complex Motion. Springer, 2004, pp. 14–29

    2004

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.