pith. sign in

arxiv: 2110.00881 · v1 · submitted 2021-10-02 · 💻 cs.CV · cs.LG

Weakly Supervised Attention-based Models Using Activation Maps for Citrus Mite and Insect Pest Classification

Edson Bollis , Helena Maia , Helio Pedrini , Sandra Avila This is my paper

Pith reviewed 2026-05-24 13:00 UTC · model grok-4.3

classification 💻 cs.CV cs.LG
keywords weakly supervised learningattention mechanismsactivation mapscitrus pest classificationmultiple instance learningpest detectionimage classification
0
0 comments X

The pith

A two-weighted activation mapping method in an attention-based two-stage network classifies tiny citrus mites and pests from class labels alone, beating prior weakly supervised approaches by at least 16 percentage points while also infering

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

The paper develops Two-Weighted Activation Mapping to generate saliency scores from class labels only and feeds those scores into an attention-based multiple instance learning stage. The resulting classifier handles the very small, noisy regions that characterize mites and insects on citrus images captured in the field. It reports higher accuracy than earlier weakly supervised baselines on both the Citrus Pest Benchmark and the larger Insect Pest dataset. The same maps also produce bounding-box locations without any location supervision during training. A reader would care because the method lowers the cost of building pest detectors by removing the need for manual bounding-box labels.

Core claim

The central claim is that the Two-Weighted Activation Mapping produces class-specific feature-map scores that, when used to guide an attention-based multiple instance learning network, deliver both higher classification accuracy on tiny pest regions and usable location estimates, all trained solely from image-level class labels.

What carries the argument

Two-Weighted Activation Mapping (TWAM), which computes saliency from class-label-driven feature maps and supplies those maps to steer attention weights inside the multiple instance learning stage.

If this is right

  • The model surpasses Attention-based Deep MIL and WILDCAT by at least 16 percentage points on both the Citrus Pest Benchmark and Insect Pest datasets.
  • Bounding-box locations for salient insects are produced at test time without any location labels seen during training.
  • The two-stage pipeline (TWAM followed by attention MIL) works on images containing multiple tiny objects against complex backgrounds.
  • Only image-level class labels are required, removing the expense of generating bounding-box annotations.

Where Pith is reading between the lines

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

  • The same label-only localization trick could be tested on other small-object domains such as weed seedlings or cell nuclei.
  • If the saliency maps prove spatially accurate when checked against held-out bounding boxes, the method supplies cheap pseudo-labels for fully supervised detectors.
  • The reported gains might shrink if future baselines adopt identical training protocols rather than published numbers.

Load-bearing premise

The class-label activation maps reliably mark the tiny mite locations rather than latching onto background texture or noise.

What would settle it

Retraining the compared Attention-based Deep MIL and WILDCAT baselines on the exact same data splits, augmentations, and optimization schedule yields accuracy within a few points of the proposed model.

Figures

Figures reproduced from arXiv: 2110.00881 by Edson Bollis, Helena Maia, Helio Pedrini, Sandra Avila.

Figure 1
Figure 1. Figure 1: Patches from CPB mites: each capture uses 60 [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 3
Figure 3. Figure 3: We take advantage of the attention-based [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figure 3
Figure 3. Figure 3: (a) Attention-based Multiple Instance Learning Guided by Saliency Maps (Attention-based MIL-Guided) consists of four steps. In Step 1, [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: RGB image transformation. (a) RGB encoded image, [PITH_FULL_IMAGE:figures/full_fig_p009_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: The effects of using: (a) removal of noisy images and (b) dropout in Bag Model training. Acronyms: ‘Atten.’ models trained using the attention-based activation map proposed approach (Two￾WAM); ‘Drop.’ models trained with dropout; ‘NCPB’ experiments trained with only images or instances from NCPB; and ‘CPB’ experi￾ments trained with original images. consider CPB or NCPB images for training (‘NCPB Validation… view at source ↗
Figure 6
Figure 6. Figure 6: b presents the results concerning the Bag Model fine-tuning in the Instance Model training. The fine-tuning does not improve the classification perfor￾mance using Two-WAM instances, but it improves us￾ing Grad-CAM instances. The fine-tuning strategy’s best result is 92.8% accuracy and 92.2% F1-score in CPB validation set, and 92.9% accuracy and 91.8% F1- score on the NCPB validation set. We achieved the be… view at source ↗
Figure 7
Figure 7. Figure 7: (a) CPB sample images. (b) Attention-based MIL [PITH_FULL_IMAGE:figures/full_fig_p013_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: (a) IP102 sample images. (b) MIL-Guided saliency maps produced by Grad-CAM. (c) Attention-based MIL-Guided saliency maps based [PITH_FULL_IMAGE:figures/full_fig_p015_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: (a) IP102 sample images [5]. (b) Attention-based MIL [PITH_FULL_IMAGE:figures/full_fig_p016_9.png] view at source ↗
read the original abstract

Citrus juices and fruits are commodities with great economic potential in the international market, but productivity losses caused by mites and other pests are still far from being a good mark. Despite the integrated pest mechanical aspect, only a few works on automatic classification have handled images with orange mite characteristics, which means tiny and noisy regions of interest. On the computational side, attention-based models have gained prominence in deep learning research, and, along with weakly supervised learning algorithms, they have improved tasks performed with some label restrictions. In agronomic research of pests and diseases, these techniques can improve classification performance while pointing out the location of mites and insects without specific labels, reducing deep learning development costs related to generating bounding boxes. In this context, this work proposes an attention-based activation map approach developed to improve the classification of tiny regions called Two-Weighted Activation Mapping, which also produces locations using feature map scores learned from class labels. We apply our method in a two-stage network process called Attention-based Multiple Instance Learning Guided by Saliency Maps. We analyze the proposed approach in two challenging datasets, the Citrus Pest Benchmark, which was captured directly in the field using magnifying glasses, and the Insect Pest, a large pest image benchmark. In addition, we evaluate and compare our models with weakly supervised methods, such as Attention-based Deep MIL and WILDCAT. The results show that our classifier is superior to literature methods that use tiny regions in their classification tasks, surpassing them in all scenarios by at least 16 percentage points. Moreover, our approach infers bounding box locations for salient insects, even training without any location labels.

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 paper proposes Two-Weighted Activation Mapping (TWAM) within an Attention-based Multiple Instance Learning pipeline guided by saliency maps for weakly supervised classification of tiny mite and insect regions. It evaluates the approach on the Citrus Pest Benchmark (field-captured images) and Insect Pest dataset, claiming at least 16 percentage point gains over Attention-based Deep MIL and WILDCAT while also producing bounding-box localizations from class labels alone.

Significance. If the performance margins are shown to arise from the proposed components rather than training-protocol differences, the work would demonstrate a practical route to localization without bounding-box supervision for small-object agronomic tasks. The emphasis on field-captured noisy data and label-efficient training aligns with real deployment constraints.

major comments (3)
  1. [Abstract / Experimental results] Abstract and experimental results: the headline claim of 'surpassing them in all scenarios by at least 16 percentage points' is load-bearing, yet the manuscript supplies no statement that the baselines were re-trained under identical data splits, augmentation, optimizer schedules, or hyper-parameters on the Citrus Pest Benchmark; any deviation can produce large deltas on small noisy datasets.
  2. [Results] Results section: no ablation is reported that isolates the contribution of the two weights in TWAM, the saliency-map guidance, or the two-stage MIL pipeline from other implementation choices; without such controls the attribution of the reported gains remains unverified.
  3. [Abstract / Localization discussion] Abstract and localization discussion: the claim that the model 'infers bounding box locations for salient insects' is presented without any quantitative localization metric (IoU, precision-recall on inferred boxes, or comparison to ground-truth boxes) or protocol for converting activation maps to boxes.
minor comments (1)
  1. [Abstract] Abstract: 'integrated pest mechanical aspect' appears to be a phrasing error; consider 'integrated pest management aspect'.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive feedback. We address each major comment below and indicate the revisions planned for the next version of the manuscript.

read point-by-point responses

  1. Referee: [Abstract / Experimental results] Abstract and experimental results: the headline claim of 'surpassing them in all scenarios by at least 16 percentage points' is load-bearing, yet the manuscript supplies no statement that the baselines were re-trained under identical data splits, augmentation, optimizer schedules, or hyper-parameters on the Citrus Pest Benchmark; any deviation can produce large deltas on small noisy datasets.

    Authors: We acknowledge that the manuscript does not explicitly confirm identical re-training of the baselines. To ensure a fair comparison, we will re-train Attention-based Deep MIL and WILDCAT using the exact same data splits, augmentations, optimizer, and hyper-parameter schedules as our method on the Citrus Pest Benchmark and report the updated results in the revised experimental section. revision: yes

  2. Referee: [Results] Results section: no ablation is reported that isolates the contribution of the two weights in TWAM, the saliency-map guidance, or the two-stage MIL pipeline from other implementation choices; without such controls the attribution of the reported gains remains unverified.

    Authors: We agree that the absence of targeted ablations leaves the source of the gains unclear. In the revised manuscript we will add ablation experiments that successively remove the two weights in TWAM, the saliency-map guidance term, and the two-stage training procedure while keeping all other implementation details fixed, thereby isolating their individual contributions. revision: yes

  3. Referee: [Abstract / Localization discussion] Abstract and localization discussion: the claim that the model 'infers bounding box locations for salient insects' is presented without any quantitative localization metric (IoU, precision-recall on inferred boxes, or comparison to ground-truth boxes) or protocol for converting activation maps to boxes.

    Authors: The datasets used are weakly supervised and contain no bounding-box annotations, so direct IoU or precision-recall against ground truth is not possible. We will nevertheless add an explicit description of the activation-to-box conversion protocol (thresholding and connected-component extraction) together with qualitative localization examples and any feasible proxy metrics. The abstract and discussion will be revised to accurately reflect these limitations and the added protocol. revision: partial

Circularity Check

0 steps flagged

No significant circularity; empirical benchmark comparisons are self-contained

full rationale

The paper proposes the TWAM activation mapping method and a two-stage MIL-guided architecture, then evaluates them via standard training and accuracy reporting on the Citrus Pest Benchmark and Insect Pest datasets. Superiority is asserted through direct numerical comparison to Attention-based Deep MIL and WILDCAT on the same benchmarks. No equations, parameters, or predictions are defined in terms of the target quantities themselves, and no load-bearing step reduces by construction to a fit or self-citation. The central claims rest on external experimental outcomes rather than tautological re-labeling of inputs.

Axiom & Free-Parameter Ledger

2 free parameters · 0 axioms · 0 invented entities

Abstract-only review yields limited visibility into exact hyperparameters; the method necessarily introduces learned weights for the activation maps and multiple training-stage thresholds that function as free parameters.

free parameters (2)
  • two weights in TWAM
    The two weighting coefficients that combine feature maps are learned or chosen to produce the final activation map.
  • saliency and MIL stage thresholds
    Decision thresholds that separate candidate regions in the two-stage pipeline are not specified and must be set.

pith-pipeline@v0.9.0 · 5832 in / 1156 out tokens · 24081 ms · 2026-05-24T13:00:41.375815+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

55 extracted references · 55 canonical work pages

  1. [1]

    M. F. Neves, V . G. Trombin, V . N. Marques, L. F. Martinez, Global orange juice market: a 16-year summary and opportu- nities for creating value, Tropical Plant Pathology 45 (2020) 166–174

    work page 2020

  2. [2]

    T. H. Spreen, Z. Gao, W. Fernandes, M. L. Zansler, Global eco- nomics and marketing of citrus products, Elsevier Inc., (2020)

    work page 2020

  3. [3]

    Bassanezi, A

    R. Bassanezi, A. Czermainski, F. Laranjeira, A. Moreira, P. Ribeiro, E. Krainski, L. Amorim, Spatial patterns of the Citrus leprosis virus and its associated mite vector in systems without intervention, Plant Pathology 68 (2019) 85–93

    work page 2019

  4. [4]

    de Carvalho, E

    S. de Carvalho, E. Girardi, F. Mour ˜ao F., R. Ferrarezi, H. Co- letta F., Advances in citrus propagation in Brazil, in: Revista Brasileira de Fruticultura, 6, (2019), pp. 1–36

    work page 2019

  5. [5]

    X. Wu, C. Zhan, Y .-K. Lai, M.-M. Cheng, J. Yang, IP102: A Large-Scale Benchmark Dataset for Insect Pest Recogni- tion, Computer Vision and Pattern Recognition (CVPR) (2019) 8787–8796

    work page 2019

  6. [6]

    H. Pei, K. Liu, X. Zhao, A. A. Yahya, Enhancing aphid detection framework based on ORB and convolutional neural networks, Scientific Reports 10 (2020) 1–15

    work page 2020

  7. [7]

    R. Wang, L. Liu, C. Xie, P. Yang, R. Li, M. Zhou, Agripest: A large-scale domain-specific benchmark dataset for practical agricultural pest detection in the wild, Sensors 21 (2021) 1–15

    work page 2021

  8. [8]

    Bollis, H

    E. Bollis, H. Pedrini, S. Avila, Weakly supervised learning guided by activation mapping applied to a novel citrus pest benchmark, Computer Vision and Pattern Recognition Work- shops (CVPRW) (2020) 310–319

    work page 2020

  9. [9]

    B. Zhou, A. Khosla, A. Lapedriza, A. Oliva, A. Torralba, Learn- ing Deep Features for Discriminative Localization, Computer Vision and Pattern Recognition (CVPR) (2016) 2921–2929

    work page 2016

  10. [10]

    T. G. Dietterich, R. H. Lathrop, T. Lozano-P ´erez, Solving the multiple instance problem with axis-parallel rectangles, Artifi- cial Intelligence 89 (1997) 31–71

    work page 1997

  11. [11]

    Durand, T

    T. Durand, T. Mordan, N. Thome, M. Cord, WILDCAT: Weakly supervised learning of deep convnets for image classification, pointwise localization and segmentation, Computer Vision and Pattern Recognition (CVPR) (2017) 5957–5966

    work page 2017

  12. [12]

    M. Ilse, J. M. Tomczak, M. Welling, Attention-based deep mul- tiple instance learning, International Conference on Machine Learning (ICML) 5 (2018) 3376–3391

    work page 2018

  13. [13]

    H. Chen, Q. Hu, B. Zhai, H. Chen, K. Liu, A robust weakly supervised learning of deep conv-nets for surface defect inspec- tion, Neural Computing and Applications (2020) 1–16

    work page 2020

  14. [14]

    Yeh, M.-H

    C.-H. Yeh, M.-H. Lin, P.-C. Chang, L.-W. Kang, Enhanced vi- sual attention-guided deep neural networks for image classifica- tion, IEEE Access 8 (2020) 163447–163457

    work page 2020

  15. [15]

    S. Woo, J. Park, J. Lee, I. S. Kweon, CBAM: Convolutional block attention module, in: European Conference on Computer Vision, 2018, pp. 3–19

    work page 2018

  16. [16]

    Y . Shen, N. Wu, J. Phang, J. Park, G. Kim, L. Moy, K. Cho, K. J. Geras, Globally-Aware Multiple Instance Classifier for Breast Cancer Screening, Lecture Notes in Computer Science (LNCS) 11861 (2019) 18–26

    work page 2019

  17. [17]

    Krizhevsky, I

    A. Krizhevsky, I. Sutskever, H. Geo ffrey E., ImageNet Classifi- cation with Deep Convolutional Neural Networks, Advances in Neural Information Processing Systems (NIPS) (2012) 1–9

    work page 2012

  18. [18]

    Szegedy, W

    C. Szegedy, W. Liu, Y . Jia, P. Sermanet, S. Reed, D. Anguelov, D. Erhan, V . Vanhoucke, A. Rabinovich, Going deeper with convolutions, in: Computer Vision and Pattern Recognition (CVPR), (2015), pp. 1–9

    work page 2015

  19. [19]

    Elliott, J

    N. Elliott, J. Farrell, A. Gutierrez, C. van Lenteren, M. Walton, S. Wratten, Integrated Pest Management, Springer Science & Business Media, (1995)

    work page 1995

  20. [20]

    Smith, H

    R. Smith, H. Reynolds, Principles, definitions and scope of in- tegrated pest control, in: FAO Symposium on Integrated Pest Control, (1966), pp. 11–17

    work page 1966

  21. [21]

    Morgan, U

    K. Morgan, U. Albrecht, F. Alferez, O. Batuman, et al., Florida Citrus Production Guide, Technical Report, Institute of Food and Agricultural Sciences, University of Florida, 2020

    work page 2020

  22. [22]

    Z. Wang, W. Gong, W. Li, A dynamic feature weighting method for mangrove pests image classification with heavy-tailed distri- butions, International Conference Proceeding Series (2020)

    work page 2020

  23. [23]

    J. Lu, J. Hu, G. Zhao, F. Mei, C. Zhang, An in-field automatic wheat disease diagnosis system, Computers and Electronics in Agriculture 142 (2017) 369–379

    work page 2017

  24. [24]

    Q. H. Cap, H. Uga, S. Kagiwada, H. Iyatomi, LeafGAN: An Ef- fective Data Augmentation Method for Practical Plant Disease Diagnosis, Transactions on Automation Science and Engineer- ing (2020) 1–10

    work page 2020

  25. [25]

    Bastianel, J

    M. Bastianel, J. Freitas-Ast ´ua, E. W. Kitajima, M. A. Machado, The citrus leprosis pathosystem, Summa Phytopathologica 32 (2006) 211–220. 17

    work page 2006

  26. [26]

    Z. H. Zhou, A brief introduction to weakly supervised learning, National Science Review 5 (2018) 44–53

    work page 2018

  27. [27]

    M. A. Carbonneau, V . Cheplygina, E. Granger, G. Gagnon, Mul- tiple instance learning: A survey of problem characteristics and applications, Pattern Recognition 77 (2018) 329–353

    work page 2018

  28. [28]

    Bahdanau, K

    D. Bahdanau, K. Cho, Y . Bengio, Neural machine translation by jointly learning to align and translate, in: Y . Bengio, Y . Le- Cun (Eds.), International Conference on Learning Representa- tions (ICLR), (2015), pp. 1–11

    work page 2015

  29. [29]

    Ra ffel, D

    C. Ra ffel, D. Ellis, Feed-forward networks with attention can solve some long-term memory problems, in: International Con- ference on Learning Representations (ICLR), 2016, pp. 1–6

    work page 2016

  30. [30]

    Chaudhari, G

    S. Chaudhari, G. Polatkan, R. Ramanath, V . Mithal, An attentive survey of attention models, arXiv 37 (2019)

    work page 2019

  31. [31]

    Borji, M.-M

    A. Borji, M.-M. Cheng, Q. Hou, H. Jiang, J. Li, Salient object detection: A survey, Computational Visual Media (2019) 117– 150

    work page 2019

  32. [32]

    J. Choe, S. J. Oh, S. Lee, S. Chun, Z. Akata, H. Shim, Evalu- ating weakly supervised object localization methods right, in: Computer Vision and Pattern Recognition (CVPR), 2020, pp. 3130–3139

    work page 2020

  33. [33]

    Adiga V ., J

    S. Adiga V ., J. Dolz, H. Lombaert, Manifold-driven attention maps for weakly supervised segmentation, arXiv:2004.03046 (2020)

    work page arXiv 2004

  34. [34]

    J. Rony, S. Belharbi, J. Dolz, I. B. Ayed, L. McCa ffrey, E. Granger, Deep weakly-supervised learning methods for classification and localization in histology images: a survey, arXiv:1909.03354v2 (2019)

    work page arXiv 1909

  35. [35]

    W. Wang, Q. Lai, H. Fu, J. Shen, H. Ling, R. Yang, Salient Ob- ject Detection in the Deep Learning Era: An In-Depth Survey, Pattern Analysis and Machine Intelligence (2019) 1–20

    work page 2019

  36. [36]

    A. d. S. Correia, E. L. Colombini, Attention, please! A survey of Neural Attention Models in Deep Learning, arXiv:2103.16775v1 (2021)

    work page arXiv 2021

  37. [37]

    Selvaraju, M

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

    work page 2017

  38. [38]

    W. S. Kim, D. H. Lee, Y . J. Kim, Machine vision-based auto- matic disease symptom detection of onion downy mildew, Com- puters and Electronics in Agriculture 168 (2020) 105099

    work page 2020

  39. [39]

    Simonyan, A

    K. Simonyan, A. Zisserman, Very deep convolutional networks for large-scale image recognition, in: International Conference on Learning Representations (ICLR), (2015), pp. 7–9

    work page 2015

  40. [40]

    Y . Chen, X. Zhang, Z. Chen, M. Song, J. Wang, Fine-grained classification of fly species in the natural environment based on deep convolutional neural network, Computers in Biology and Medicine 135 (2021) 104655

    work page 2021

  41. [41]

    E. A. Lins, J. P. M. Rodriguez, S. I. Scoloski, J. Pivato, M. B. Lima, J. M. C. Fernandes, P. R. V . da Silva Pereira, D. Lau, R. Rieder, A method for counting and classifying aphids using computer vision, Computers and Electronics in Agriculture 169 (2020) 105200

    work page 2020

  42. [42]

    Y . Wu, L. Xu, Crop organ segmentation and disease identifica- tion based on weakly supervised deep neural network, Agron- omy 9 (2019)

    work page 2019

  43. [43]

    M. Tan, Q. Le, E fficientNet: Rethinking model scaling for con- volutional neural networks, in: International Conference on Ma- chine Learning (ICML), (2019), pp. 6105–6114

    work page 2019

  44. [44]

    Howard, M

    A. Howard, M. Sandler, B. Chen, W. Wang, L. Chen, M. Tan, G. Chu, V . Vasudevan, Y . Zhu, R. Pang, H. Adam, Q. Le, Searching for MobileNetV3, in: International Conference on Computer Vision (ICCV), (2019), pp. 1314–1324

    work page 2019

  45. [45]

    J. Hu, L. Shen, S. Albanie, G. Sun, E. Wu, Squeeze-and- Excitation Networks, Transactions on Pattern Analysis and Ma- chine Intelligence 42 (2020) 2011–2023

    work page 2020

  46. [46]

    L. Liu, R. Wang, C. Xie, P. Yang, F. Wang, S. Sudirman, W. Liu, PestNet: An End-to-End Deep Learning Approach for Large- Scale Multi-Class Pest Detection and Classification, IEEE Ac- cess 7 (2019) 45301–45312

    work page 2019

  47. [47]

    F. Wang, R. Wang, C. Xie, P. Yang, L. Liu, Fusing multi-scale context-aware information representation for automatic in-field pest detection and recognition, Computers and Electronics in Agriculture 169 (2020) 105222

    work page 2020

  48. [48]

    W. Zeng, M. Li, Crop leaf disease recognition based on Self- Attention convolutional neural network, Computers and Elec- tronics in Agriculture 172 (2020) 105341

    work page 2020

  49. [49]

    L. Deng, Y . Wang, Z. Han, R. Yu, Research on insect pest im- age detection and recognition based on bio-inspired methods, Biosystems Engineering 169 (2018) 139–148

    work page 2018

  50. [50]

    Nanni, G

    L. Nanni, G. Maguolo, F. Pancino, Insect pest image detection and recognition based on bio-inspired methods, Ecological In- formatics 57 (2020)

    work page 2020

  51. [51]

    Costello, Mites, in: P

    M. Costello, Mites, in: P. Christensen (Ed.), Raisin production manual, UCANR Publications, (2000), pp. 187–190

    work page 2000

  52. [52]

    Robbins, S

    H. Robbins, S. Monro, A stochastic approximation method, The annals of mathematical statistics (1951) 400–407

    work page 1951

  53. [53]

    Kiefer, J

    J. Kiefer, J. Wolfowitz, Stochastic estimation of the maximum of a regression function, The Annals of Mathematical Statistics (1952) 462–466

    work page 1952

  54. [54]

    LeCun, L

    Y . LeCun, L. Bottou, Y . Bengio, P. Ha ffner, Gradient-based learning applied to document recognition, IEEE Access 86 (1998) 2278–2323

    work page 1998

  55. [55]

    S. Wu, S. Zhong, Y . Liu, Deep residual learning for image recognition, Multimedia Tools and Applications (2017) 1–17. 18

    work page 2017