Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Detection of Spider Mites on Labrador Beans through Machine Learning Approaches Using Custom Datasets

Violet Liu, Jason Chen, Ans Qureshi, Mahla Nejati

TL;DR

This paper tackles early detection of spider mites on labrador beans by building an RGBN dataset from real greenhouse conditions and evaluating a two-stage pipeline that couples YOLOv8-based leaf segmentation with a sequential CNN classifier. It demonstrates that RGBN input generally improves classification performance and that a two-stage approach effectively handles partially labeled data, outperforming single-stage strategies on occluded datasets. Key contributions include novel RGBN data collection, grid-based segmentation to preserve leaf detail, and an analysis of channel fusion strategies and transfer learning, with the sequential CNN achieving $90.62\%$ validation accuracy on RGBN. The findings suggest that incorporating NIR data can enhance early disease detection and that modular, stage-wise processing improves robustness in practical agricultural settings, though larger datasets and more powerful hardware are needed for broader deployment.

Abstract

Amidst growing food production demands, early plant disease detection is essential to safeguard crops; this study proposes a visual machine learning approach for plant disease detection, harnessing RGB and NIR data collected in real-world conditions through a JAI FS-1600D-10GE camera to build an RGBN dataset. A two-stage early plant disease detection model with YOLOv8 and a sequential CNN was used to train on a dataset with partial labels, which showed a 3.6% increase in mAP compared to a single-stage end-to-end segmentation model. The sequential CNN model achieved 90.62% validation accuracy utilising RGBN data. An average of 6.25% validation accuracy increase is found using RGBN in classification compared to RGB using ResNet15 and the sequential CNN models. Further research and dataset improvements are needed to meet food production demands.

Detection of Spider Mites on Labrador Beans through Machine Learning Approaches Using Custom Datasets

TL;DR

This paper tackles early detection of spider mites on labrador beans by building an RGBN dataset from real greenhouse conditions and evaluating a two-stage pipeline that couples YOLOv8-based leaf segmentation with a sequential CNN classifier. It demonstrates that RGBN input generally improves classification performance and that a two-stage approach effectively handles partially labeled data, outperforming single-stage strategies on occluded datasets. Key contributions include novel RGBN data collection, grid-based segmentation to preserve leaf detail, and an analysis of channel fusion strategies and transfer learning, with the sequential CNN achieving validation accuracy on RGBN. The findings suggest that incorporating NIR data can enhance early disease detection and that modular, stage-wise processing improves robustness in practical agricultural settings, though larger datasets and more powerful hardware are needed for broader deployment.

Abstract

Amidst growing food production demands, early plant disease detection is essential to safeguard crops; this study proposes a visual machine learning approach for plant disease detection, harnessing RGB and NIR data collected in real-world conditions through a JAI FS-1600D-10GE camera to build an RGBN dataset. A two-stage early plant disease detection model with YOLOv8 and a sequential CNN was used to train on a dataset with partial labels, which showed a 3.6% increase in mAP compared to a single-stage end-to-end segmentation model. The sequential CNN model achieved 90.62% validation accuracy utilising RGBN data. An average of 6.25% validation accuracy increase is found using RGBN in classification compared to RGB using ResNet15 and the sequential CNN models. Further research and dataset improvements are needed to meet food production demands.
Paper Structure (27 sections, 7 figures, 6 tables)

This paper contains 27 sections, 7 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Utilising RGB and NIR Channels in Data
  4. Dataset exploration
  5. State-of-the-art Segmentation Models
  6. State-of-the-art Classification Models
  7. Methodology
  8. Two-stage Model
  9. Single-stage Model
  10. Segmentation
  11. Channel Fusing Techniques
  12. Substitution technique
  13. Addition Technique
  14. Classification
  15. Results
  16. ...and 12 more sections

Figures (7)

  • Figure 1: Comparison of the Single-Stage Model (a) and the Two-Stage Model (b).
  • Figure 2: Dataset conversion for segmentation model.
  • Figure 3: Dataset conversion for classification model.
  • Figure 4: Removal of leaf class to obtain a fully labelled dataset.
  • Figure 5: Visual representation of camera distance to plants.
  • ...and 2 more figures