pith. sign in

arxiv: 2604.13835 · v1 · submitted 2026-04-15 · 💻 cs.CV

A Resource-Efficient Hybrid CNN-LSTM network for image-based bean leaf disease classification

Hye Jin Rhee , Joseph Damilola Akinyemi This is my paper

Pith reviewed 2026-05-10 13:01 UTC · model grok-4.3

classification 💻 cs.CV
keywords bean leaf diseaseCNN-LSTM hybridimage classificationresource efficientplant pathologyimage augmentationlightweight modelagricultural AI
0
0 comments X

The pith

A hybrid CNN-LSTM architecture classifies bean leaf diseases at 94.38% accuracy with a model size of only 1.86 MB.

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

The paper introduces a lightweight hybrid system that combines convolutional neural networks with long short-term memory layers for identifying diseases in bean leaves from photographs. Standard CNNs struggle with capturing relationships across distant parts of an image due to pooling operations, but inserting an LSTM layer allows the model to treat feature maps as sequences and learn those dependencies. This design delivers high accuracy on the ibean dataset while shrinking the model to a size suitable for mobile or embedded devices in farming. The work also compares different ways of augmenting training images and finds that carefully chosen transformations preserve disease features better than broad random changes. Such a system could support automated crop monitoring tools that run locally without constant internet access or powerful hardware.

Core claim

By integrating an LSTM layer to model the spatial-sequential relationships within feature maps, the hybrid architecture achieves a 94.38% accuracy while maintaining an exceptionally small footprint of 1.86 MB, a 70% reduction in size compared to traditional CNN-based systems, and state-of-the-art F1 scores of 99.22% with EfficientNet-B7+LSTM on the ibean dataset.

What carries the argument

LSTM layer integrated after CNN feature extraction to model spatial-sequential relationships within the feature maps.

If this is right

  • Enables real-time agricultural decision support in resource-constrained environments.
  • Tailored image augmentations outperform generic combinations for preserving diagnostic patterns.
  • Small model size supports deployment on portable devices for on-site diagnosis.
  • EfficientNet-B7 combined with LSTM reaches top F1 performance on bean leaf tasks.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The hybrid structure may improve efficiency for image-based disease classification in other crops.
  • Domain-specific augmentation choices indicate that general augmentation tools often fall short in plant pathology.
  • Memory reduction could make AI diagnosis accessible to farms with basic hardware.
  • Further tests across seasons and bean varieties would check reliability outside the original dataset.

Load-bearing premise

The ibean dataset together with the selected image augmentations sufficiently represent real-world variability in bean leaf appearance, lighting, and disease presentation.

What would settle it

A significant drop in classification accuracy when the model is tested on a fresh set of bean leaf images collected from different locations or under new lighting and growth conditions.

Figures

Figures reproduced from arXiv: 2604.13835 by Hye Jin Rhee, Joseph Damilola Akinyemi.

Figure 1
Figure 1. Figure 1: Three classes of bean leaf images in the ibean dataset [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 3
Figure 3. Figure 3: Custom architecture of lightweight models related patterns, this hybrid approach allows the model to contextualise localised symptoms within the broader leaf geometry, providing a more robust feature representation than isolated fully-connected layers [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Activation map by custom lightweight model [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 6
Figure 6. Figure 6: Training Bean-CNN-LSTM on the original set [PITH_FULL_IMAGE:figures/full_fig_p006_6.png] view at source ↗
Figure 5
Figure 5. Figure 5: Training Bean-CNN on the original set MCC in Bean-CNN-LSTM implies that this model made fewer false predictions than the Bean-CNN models. We also analyse the average results from all 30 training runs for a more general tendency [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗
Figure 8
Figure 8. Figure 8: The average test accuracy on each training set: The red line indicates the median value, and a box represents the distribution between 25 and 75 percentiles. Small hollow circles indicate outliers [PITH_FULL_IMAGE:figures/full_fig_p007_8.png] view at source ↗
Figure 10
Figure 10. Figure 10: Confusion matrix: The best Bean-CNN-LSTM model In the second model, the classification layer is replaced with an LSTM layer and a linear classification layer, and a Dropout layer (50%) in between them. We refer to the former model as the EfficientNetB7+FC model and the latter as the EfficientNetB7+LSTM model. For best results, the entire architecture was trained, but the EfficientNet backbone was unfrozen… view at source ↗
Figure 9
Figure 9. Figure 9: Box-plot representation: The average performance of our lightweight custom models (Bean-CNN and Bean-CNN￾LSTM) from 30 training runs comparison across previous models, the test set remains the 128 images provided in the original dataset. To allow for some comparison between conventional Dense layers and LSTM layers, we created two separate models from EfficientNet by replacing its final classification laye… view at source ↗
Figure 11
Figure 11. Figure 11: GradCAM Feature maps. The heatmaps indicate that the model correctly prioritises the necrotic centres of the lesions rather than the leaf edges, validating the management reliability of the system. within the feature map ( [PITH_FULL_IMAGE:figures/full_fig_p010_11.png] view at source ↗
read the original abstract

Accurate and resource-efficient automated diagnosis is a cornerstone of modern agricultural expert systems. While Convolutional Neural Networks (CNNs) have established benchmarks in plant pathology, their ability to capture long-range spatial dependencies is often limited by standard pooling layers, and their high memory footprint hinders deployment on portable devices. This paper proposes a lightweight hybrid CNN-LSTM system for bean leaf disease classification. By integrating an LSTM layer to model the spatial-sequential relationships within feature maps, our hybrid architecture achieves a 94.38% accuracy while maintaining an exceptionally small footprint of 1.86 MB; a 70% reduction in size compared to traditional CNN-based systems. Furthermore, we provide a systematic evaluation of image augmentation strategies, demonstrating that tailored transformations are superior to generic combinations for maintaining the integrity of diagnostic patterns. Results on the $\textit{ibean}$ dataset confirm that the proposed system achieves state-of-the-art F1 scores of 99.22% with EfficientNet-B7+LSTM, providing a robust and scalable framework for real-time agricultural decision support in resource-constrained environments. The code and augmented datasets used in this study are publicly available on this $\href{https://github.com/HJin-R/bean_disease}{Github}$ repo.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 1 minor

Summary. The manuscript proposes a lightweight hybrid CNN-LSTM architecture for classifying bean leaf diseases on the ibean dataset. It claims that adding an LSTM layer to model spatial-sequential relationships in CNN feature maps yields 94.38% accuracy at 1.86 MB model size (70% smaller than traditional CNNs), state-of-the-art F1 scores of 99.22% when paired with EfficientNet-B7, and superior results from tailored image augmentations. Public code and augmented datasets are provided.

Significance. If the empirical claims hold after verification, this would offer a practical advance in resource-efficient models for real-time plant disease diagnosis on portable devices, addressing deployment constraints in agriculture. Public code availability supports reproducibility and potential follow-on work.

major comments (3)
  1. [Abstract] Abstract: The central claim attributes performance gains and the 1.86 MB size to the LSTM modeling 'spatial-sequential relationships within feature maps.' Feature maps are 2D (H×W×C), yet no description is given of the required reshaping, flattening (row/column/patch-wise), or projection step to produce LSTM sequences. Without this or an ablation isolating LSTM contribution versus the base CNN, the hybrid mechanism and size benefit cannot be verified as load-bearing.
  2. [Abstract] Abstract: The 70% size reduction is stated relative to 'traditional CNN-based systems' with no named baselines, their reported sizes, or calculation details (e.g., parameter count vs. memory footprint). This directly undermines the resource-efficiency claim that is central to the paper's contribution.
  3. [Abstract] Abstract: State-of-the-art F1 (99.22% with EfficientNet-B7+LSTM) and accuracy (94.38%) claims lack any mention of train-test splits, number of runs, statistical tests, or direct comparisons to other models on identical splits. These omissions make the empirical results unverifiable and affect soundness of the hybrid architecture evaluation.
minor comments (1)
  1. [Abstract] Abstract: The mention of 'systematic evaluation of image augmentation strategies' would benefit from a one-sentence summary of the key tailored transformations and their measured impact to strengthen the abstract.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive and detailed feedback. We address each major comment point by point below and indicate where revisions will be made to strengthen the manuscript.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The central claim attributes performance gains and the 1.86 MB size to the LSTM modeling 'spatial-sequential relationships within feature maps.' Feature maps are 2D (H×W×C), yet no description is given of the required reshaping, flattening (row/column/patch-wise), or projection step to produce LSTM sequences. Without this or an ablation isolating LSTM contribution versus the base CNN, the hybrid mechanism and size benefit cannot be verified as load-bearing.

    Authors: We agree that the reshaping mechanism and its contribution require explicit clarification. In the revised manuscript, we will add a precise description in the methods section explaining that the 2D feature maps are flattened row-wise into sequences (each spatial row treated as a time step) with a linear projection to match LSTM input dimensions. We will also include a new ablation study comparing the full hybrid CNN-LSTM against the base CNN without the LSTM layer, reporting accuracy, F1, and model size to isolate the LSTM's role. revision: yes

  2. Referee: [Abstract] Abstract: The 70% size reduction is stated relative to 'traditional CNN-based systems' with no named baselines, their reported sizes, or calculation details (e.g., parameter count vs. memory footprint). This directly undermines the resource-efficiency claim that is central to the paper's contribution.

    Authors: We acknowledge the need for concrete baselines and methodology. The revision will name specific traditional CNN models (ResNet50 and VGG16) used for comparison, report their sizes in MB, and detail the 70% reduction calculation based on total parameter counts converted to memory footprint (float32 precision). These will be presented in a new comparison table in the results section. revision: yes

  3. Referee: [Abstract] Abstract: State-of-the-art F1 (99.22% with EfficientNet-B7+LSTM) and accuracy (94.38%) claims lack any mention of train-test splits, number of runs, statistical tests, or direct comparisons to other models on identical splits. These omissions make the empirical results unverifiable and affect soundness of the hybrid architecture evaluation.

    Authors: We agree these details are essential for verifiability. The revised manuscript will explicitly state the train-test split ratio, report results from multiple independent runs with mean and standard deviation, and include direct comparisons to other models on identical splits. Formal statistical tests were not performed, but variance across runs will be reported to support reliability. The public code repository already enables exact reproduction of the splits and experiments. revision: partial

Circularity Check

0 steps flagged

No circularity; empirical results on external public dataset

full rationale

The paper reports direct empirical measurements of accuracy (94.38%), model size (1.86 MB), and F1 scores from training a hybrid CNN-LSTM on the ibean dataset with stated code availability. No mathematical derivations, predictions, or first-principles results are claimed that reduce to quantities defined by the authors' own fitted parameters, self-citations, or ansatzes. The architecture description and augmentation evaluation are implementation choices evaluated against external benchmarks, with no load-bearing self-citation chains or self-definitional steps present.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claims rest on the representativeness of the ibean dataset and standard deep-learning assumptions about generalization from augmented training data; no new physical or mathematical entities are introduced.

free parameters (1)
  • model hyperparameters and augmentation parameters
    Typical deep learning training choices such as learning rate, batch size, and specific augmentation strengths are fitted or selected but not enumerated in the abstract.
axioms (1)
  • domain assumption The ibean dataset provides a sufficient and unbiased benchmark for evaluating bean leaf disease classification performance.
    All accuracy and F1 claims are derived exclusively from experiments on this dataset.

pith-pipeline@v0.9.0 · 5521 in / 1275 out tokens · 48891 ms · 2026-05-10T13:01:23.083085+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

49 extracted references · 49 canonical work pages · 2 internal anchors

  1. [1]

    Pamela, D

    P. Pamela, D. Mawejje, M. Ugen, Severity of angular leaf spot and rust diseases on common beans in Central Uganda, Uganda Journal of Agricultural Sciences 14 (1) (2014). URLhttps://www.ajol.info/index.php/ujas/article/view/126189

    work page 2014

  2. [2]

    Venbrux, S

    M. Venbrux, S. Crauwels, H. Rediers, Current and emerging trends in techniques for plant pathogen detection, Frontiers in Plant Science 14 (2023).doi:10.3389/fpls.2023.1120968. URLhttps://doi.org/10.3389/fpls.2023.1120968

    work page doi:10.3389/fpls.2023.1120968 2023

  3. [3]

    Mahlein, Plant disease detection by imaging sensors–parallels and specific demands for precision agriculture and plant pheno- typing, Plant Disease 100 (2) (2016) 241–251

    A.-K. Mahlein, Plant disease detection by imaging sensors–parallels and specific demands for precision agriculture and plant pheno- typing, Plant Disease 100 (2) (2016) 241–251. doi:10.1094/ PDIS-03-15-0340-FE. URLhttps://doi.org/10.1094/PDIS-03-15-0340-FE

    work page doi:10.1094/pdis-03-15-0340-fe 2016

  4. [4]

    In: Proceedings of the 3rd International Conference on Smart Data Intelligence (ICSMDI), pp

    L. Rahunathan, D. Sivabalaselvamani, E. Elakkiya, M. Madhumitha, K. Kumaresh, Recognition of bean leaf diseases using neural net- work and machine learning techniques, in: 2023 3rd International Conference on Smart Data Intelligence (ICSMDI), 2023, pp. 520–526. doi:10.1109/ICSMDI57622.2023.00098

    work page doi:10.1109/icsmdi57622.2023.00098 2023

  5. [5]

    H. Slimani, Artificial Intelligence-based Detection of Fava Bean Rust Disease in Agricultural Settings: An Innovative Approach, International Journal of Advanced Computer Science & Applications 14 (6) (2023) 119–128.doi:doi.org/10.14569/IJACS

    work page doi:10.14569/ijacs 2023

  6. [6]

    Islam, M

    Z. Islam, M. Islam, A. Amanullah, A combined deep CNN-LSTM network for the detection of novel coronavirus (COVID-19) using X-ray images, Informatics in Medicine Unlocked 20 (2020) 100412– 100412.doi:doi.org/10.1016/j.imu.2020.100412

    work page doi:10.1016/j.imu.2020.100412 2020

  7. [7]

    Saenko, T

    J.Donahue,L.A.Hendricks,M.Rohrbach,S.Venugopalan,S.Guadar- rama, K. Saenko, T. Darrell, Long-term recurrent convolutional networks for visual recognition and description, IEEE Transactions on Pattern Analysis and Machine Intelligence 39 (4) (2017) 677–691. doi:10.1109/TPAMI.2016.2599174

    work page doi:10.1109/tpami.2016.2599174 2017

  8. [8]

    Önler, Feature fusion based artificial neural network model for disease detection of bean leaves., Electronic Research Archive 31 (5) (2023)

    E. Önler, Feature fusion based artificial neural network model for disease detection of bean leaves., Electronic Research Archive 31 (5) (2023)

    work page 2023

  9. [9]

    Elfatimi, R.Eryigit,L

    E. Elfatimi, R.Eryigit,L. Elfatimi, Beans LeafDiseases Classification Using MobileNet Models, IEEE Access 10 (2022) 9471–9482.doi: 10.1109/ACCESS.2022.3142817

    work page doi:10.1109/access.2022.3142817 2022

  10. [10]

    6105–6114

    M.Tan,Q.Le,Efficientnet:Rethinkingmodelscalingforconvolutional neural networks, in: International conference on machine learning, PMLR, 2019, pp. 6105–6114

    work page 2019

  11. [11]

    Z. Jian, Z. Wei, Support vector machine for recognition of cucumber leaf diseases, in: 2010 2nd International Conference on Advanced Computer Control, Vol. 5, 2010, pp. 264–266.doi:10.1109/ICACC. 2010.5487242

    work page doi:10.1109/icacc 2010

  12. [12]

    Y. Lu, S. Yi, N. Zeng, Y. Liu, Y. Zhang, Identification of rice diseases using deep convolutional neural networks, Neurocomputing 267 (2017) 378–384.doi:https://doi.org/10.1016/j.neucom.2017.06.023. URL https://www.sciencedirect.com/science/article/pii/ S0925231217311384

    work page doi:10.1016/j.neucom.2017.06.023 2017

  13. [13]

    Geetharamani, J

    G. Geetharamani, J. Arun Pandian, Identification of plant leaf diseases using a nine-layer deep convolutional neural network, Computers & Electrical Engineering 76 (2019) 323–338. doi:https://doi.org/10.1016/j.compeleceng.2019.04.011. URL https://www.sciencedirect.com/science/article/pii/ S0045790619300023

    work page doi:10.1016/j.compeleceng.2019.04.011 2019

  14. [14]

    S. P. Mohanty, Using Deep Learning for Image-Based Plant Disease Detection, Frontiers in Plant Science 7 (2016) 1419.doi:doi.org/10. 3389/fpls.2016.01419

    work page arXiv 2016

  15. [15]

    doi:doi.org/10.1007/s42979-023-02245-7

    M.A.Patil,M.Manohar,Plantleafdiseaseclassificationusingoptimal tuned hybrid lstm-cnn model., SN Computer Science 4 (6) (2023) 710. doi:doi.org/10.1007/s42979-023-02245-7

    work page doi:10.1007/s42979-023-02245-7 2023

  16. [16]

    E. Devi, S. Gopi, U. Padmavathi, S. R. Arumugam, S. Premnath, D.Muralitharan,Plantdiseaseclassificationusingcnn-lstmtechniques, HJ Rhee and J Akinyemi:Preprint submitted to ElsevierPage 10 of 12 Hybrid CNN-LSTM network for image-based bean leaf disease classification in: 2023 5th International Conference on Smart Systems andInventive Technology(ICSSIT),20...

    work page doi:10.1109/icssit55814 2023

  17. [17]

    M. A. Haque, C. K. Deb, P. Gole, S. Karmakar, A. Dheeraj, M. U. Din Shah, S. Dutta, M. K. P. Kumar, S. Marwaha, An enhanced vision transformer network for efficient and accurate crop disease detection, Expert Systems with Applications 283 (2025) 127743. doi:https://doi.org/10.1016/j.eswa.2025.127743. URL https://www.sciencedirect.com/science/article/pii/ ...

    work page doi:10.1016/j.eswa.2025.127743 2025

  18. [18]

    doi:https://doi.org/10.1016/j.compag.2021.106125

    A.Abade,P.A.Ferreira,F.deBarrosVidal,Plantdiseasesrecognition on images using convolutional neural networks: A systematic review, Computers and Electronics in Agriculture 185 (2021) 106125. doi:https://doi.org/10.1016/j.compag.2021.106125. URL https://www.sciencedirect.com/science/article/pii/ S0168169921001435

    work page doi:10.1016/j.compag.2021.106125 2021

  19. [19]

    1394–1399.doi:10.1109/I-SMAC61858.2024.10714612

    S.Singla,R.Gupta,DeepLearningbasedBeanLeafLesionClassifica- tion utilizing EfficientNetV2-S, in: 2024 8th International Conference on I-SMAC (IoT in Social, Mobile, Analytics and Cloud) (I-SMAC), 2024, pp. 1394–1399.doi:10.1109/I-SMAC61858.2024.10714612

    work page doi:10.1109/i-smac61858.2024.10714612 2024

  20. [20]

    Rodríguez-Lira, D.-M

    D.-C. Rodríguez-Lira, D.-M. Córdova-Esparza, J. M. Álvarez Al- varado, J.-A. Romero-González, J. Terven, J. Rodríguez-Reséndiz, Comparative Analysis of YOLO Models for Bean Leaf Disease Detection in Natural Environments, AgriEngineering 6 (4) (2024) 4585–4603.doi:10.3390/agriengineering6040262. URLhttps://www.mdpi.com/2624-7402/6/4/262

    work page doi:10.3390/agriengineering6040262 2024

  21. [21]

    Long and M

    F. Hohman, M. B. Kery, D. Ren, D. Moritz, Model Compression in Practice: Lessons Learned from Practitioners Creating On-device MachineLearningExperiences,in:ProceedingsoftheCHIConference on Human Factors in Computing Systems, CHI ’24, ACM, 2024, pp. 1–18.doi:10.1145/3613904.3642109. URLhttp://dx.doi.org/10.1145/3613904.3642109

    work page doi:10.1145/3613904.3642109 2024

  22. [22]

    H. Sun, H. Xu, B. Liu, D. He, J. He, H. Zhang, N. Geng, MEAN-SSD: A novel real-time detector for apple leaf diseases using improved light-weight convolutional neural networks, Computers and Electronics in Agriculture 189 (2021) 106379. doi:https://doi.org/10.1016/j.compag.2021.106379. URL https://www.sciencedirect.com/science/article/pii/ S0168169921003963

    work page doi:10.1016/j.compag.2021.106379 2021

  23. [23]

    Solving Current Limitations of Deep Learning Based Approaches for Plant Disease Detection,

    M.Arsenovic,M.Karanovic,S.Sladojevic,A.Anderla,D.Stefanovic, Solvingcurrentlimitationsofdeeplearningbasedapproachesforplant disease detection, Symmetry 11 (7) (2019).doi:10.3390/sym11070939. URLhttps://www.mdpi.com/2073-8994/11/7/939

    work page doi:10.3390/sym11070939 2019

  24. [24]

    S. Yang, W. Xiao, M. Zhang, S. Guo, J. Zhao, F. Shen, Image Data Augmentation for Deep Learning: A SurveyArXiv-eprint: 2204.08610 (2023)

    work page arXiv 2023

  25. [25]

    Taylor, G

    L. Taylor, G. Nitschke, Improving deep learning with generic data augmentation, in: 2018 IEEE Symposium Series on Computational Intelligence (SSCI), 2018, pp. 1542–1547.doi:10.1109/SSCI.2018. 8628742

    work page doi:10.1109/ssci.2018 2018

  26. [26]

    C.Shorten,T.M.Khoshgoftaar,Asurveyonimagedataaugmentation for deep learning, Journal of Big Data 6 (60) (2019).doi:https: //doi.org/10.1186/s40537-019-0197-0

    work page doi:10.1186/s40537-019-0197-0 2019

  27. [27]

    AI-Lab-Makerere / ibean, https://github.com/AI-Lab-Makerere/ibean/ (Jan. 2020)

    work page 2020

  28. [28]

    Metallurgy and Design of Alloys with Hierarchical Microstructures

    A.Muimba-Kankolongo, Food Crop Production by Smallholder Farmers in Southern Africa, Science Direct, 2018. URL https://www.sciencedirect.com/book/9780128143834/ food-crop-production-by-smallholder-farmers-in-southern-africa

    work page arXiv 2018

  29. [29]

    L. Deng, J. C. Platt, Ensemble deep learnig for speech recognition, Interspeech 1 (2014).doi:doi:10.21437/Interspeech.2014-433

    work page doi:10.21437/interspeech.2014-433 2014

  30. [30]

    T. N. Sainath, O. Vinyals, A. Senior, H. Sak, Convolutional, Long Short-Term Memory, fully connected Deep Neural Networks, in: 2015 IEEE International Conference on Acoustics, Speech and Signal Processing(ICASSP),2015,pp.4580–4584. doi:10.1109/ICASSP.2015. 7178838

    work page doi:10.1109/icassp.2015 2015

  31. [31]

    Ercolano, S

    G. Ercolano, S. Rossi, Combining CNN and LSTM for activity of daily living recognition with a 3D matrix skeleton representation, IntelligentServiceRobotics14(2021)175–185. doi:doi.org/10.1007/ s11370-021-00358-7

    work page 2021

  32. [32]

    Very Deep Convolutional Networks for Large-Scale Image Recognition

    K. Simonyan, A. Zisserman, Very deep convolutional networks for large-scale image recognition, in: Y. Bengio, Y. LeCun (Eds.), 3rd International Conference on Learning Representations, ICLR 2015, San Diego, CA, USA, May 7-9, 2015, Conference Track Proceedings, 2015. URLhttp://arxiv.org/abs/1409.1556

    work page internal anchor Pith review Pith/arXiv arXiv 2015

  33. [33]

    Zhang, N

    X. Zhang, N. Han, J. Zhang, Comparative analysis of vgg, resnet, and googlenet architectures evaluating performance, computational efficiency, and convergence rates, Applied and Computational Engi- neering 44 (2024) 172–181

    work page 2024

  34. [34]

    M. S. Nixon, A. A. Aguado, Feature Extraction and Image Processing for Computer Vision,, Academic Press, London, 2020

    work page 2020

  35. [35]

    K. He, X. Zhang, S. Ren, J. Sun, Delving Deep into Rectifiers: Surpassing Human-Level Performance on ImageNet Classification, in: 2015 IEEE International Conference on Computer Vision (ICCV), 2015, pp. 1026–1034.doi:10.1109/ICCV.2015.123

    work page doi:10.1109/iccv.2015.123 2015

  36. [36]

    Long short- term memory,

    S. Hochreiter, J. Schmidhuber, Long short-term memory, Neural Computation 9 (8) (1997) 1735–1780.arXiv:https://direct.mit. edu/neco/article-pdf/9/8/1735/813796/neco.1997.9.8.1735.pdf, doi: 10.1162/neco.1997.9.8.1735. URLhttps://doi.org/10.1162/neco.1997.9.8.1735

    work page doi:10.1162/neco.1997.9.8.1735 1997

  37. [37]

    A.Tharwat,Classificationassessmentmethods,AppliedComputing& Informatics 17 (1) (2021) 168–192.doi:doi.org/10.1016/j.aci.2018. 08.003

    work page doi:10.1016/j.aci.2018 2021

  38. [38]

    Chicco, G

    D. Chicco, G. Jurman, The advantages of the Matthews correlation coefficient (MCC) over F1 score and accuracy in binary classification evaluation., BMC Genomics 21 (1) (Jan. 2020).doi:doi:10.1186/ s12864-019-6413-7

    work page 2020

  39. [39]

    R. R. Selvaraju, M. Cogswell, A. Das, R. Vedantam, D. Parikh, D. Batra, Grad-CAM: Visual Explanations from Deep Networks via Gradient-Based Localization, in: 2017 IEEE International Conference on Computer Vision (ICCV), 2017, pp. 618–626.doi:10.1109/ICCV. 2017.74

    work page doi:10.1109/iccv 2017

  40. [40]

    Goodfellow, Y

    I. Goodfellow, Y. Bengio, A. Courville, Deep Learning, MIT Press, 2016

    work page 2016

  41. [41]

    D. P. Kingma, J. Ba, Adam: A method for stochastic optimization, arXiv preprint arXiv:1412.6980 (2014)

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  42. [42]

    S. H. Abed, A. S. Al-Waisy, H. J. Mohammed, S. Al-Fahdawi, A modern deep learning framework in robot vision for automated bean leaves diseases detection, International Journal of Intelligent Robotics and Applications 5 (2) (2021) 235–251

    work page 2021

  43. [43]

    Singh, A

    V. Singh, A. Chug, A. P. Singh, Classification of beans leaf diseases using fine tuned cnn model, Procedia Computer Science 218 (2023) 348–356

    work page 2023

  44. [44]

    Sunyoto, D

    A. Sunyoto, D. Ariatmanto, et al., Innovative solutions for bean leaf disease detection using deep learning, in: 2024 IEEE International Conference on Artificial Intelligence and Mechatronics Systems (AIMS), IEEE, 2024, pp. 1–5

    work page 2024

  45. [45]

    E. Jain, A. Aneja, Automated detection and classification of bean leaf diseases using inceptionv3: A deep learning approach, in: 2025 International Conference on Electronics and Renewable Systems (ICEARS), 2025, pp. 1890–1895. doi:10.1109/ICEARS64219.2025. 10941547

    work page doi:10.1109/icears64219.2025 2025

  46. [46]

    Karthik, R

    R. Karthik, R. Aswin, K. S. Geetha, K. Suganthi, An explainable deep learning network with transformer and custom cnn for bean leaf disease classification, IEEE Access 13 (2025) 38562–38573. doi:10.1109/ACCESS.2025.3546017

    work page doi:10.1109/access.2025.3546017 2025

  47. [47]

    Efficient attention: Attention with linear complexities

    K. Kahatapitiya, R. Rodrigo, Exploiting the Redundancy in Con- volutional Filters for Parameter Reduction, in: 2021 IEEE Winter Conference on Applications of Computer Vision (WACV), 2021, pp. 1409–1419.doi:10.1109/WACV48630.2021.00145

    work page doi:10.1109/wacv48630.2021.00145 2021

  48. [48]

    J. G. A. Barbedo, A review on the main challenges in automatic plant disease identification based on visible range images, Biosystems Engineering 144 (2016) 52–60. doi:https://doi.org/10.1016/j.biosystemseng.2016.01.017. URL https://www.sciencedirect.com/science/article/pii/ HJ Rhee and J Akinyemi:Preprint submitted to ElsevierPage 11 of 12 Hybrid CNN-LST...

    work page doi:10.1016/j.biosystemseng.2016.01.017 2016

  49. [49]

    G. Fenu, F. M. Malloci, DiaMOS Plant: A Dataset for Diagnosis and Monitoring Plant Disease, Agronomy 11 (11) (2021). doi: 10.3390/agronomy11112107. URLhttps://www.mdpi.com/2073-4395/11/11/2107 HJ Rhee and J Akinyemi:Preprint submitted to ElsevierPage 12 of 12

    work page doi:10.3390/agronomy11112107 2021