Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Forestpest-YOLO: A High-Performance Detection Framework for Small Forestry Pests

Aoduo Li, Peikai Lin, Jiancheng Li, Zhen Zhang, Shiting Wu, Zexiao Liang, Zhifa Jiang

TL;DR

The paper tackles detecting extremely small forestry pests in complex UAV imagery, where targets are highly occluded and data are imbalanced. It introduces Forestpest-YOLO, a YOLOv8-based detector augmented with SPD-Conv for lossless downsampling, CSPOK for robust cross-scale fusion, and VarifocalLoss to focus learning on hard/high-quality samples. On the self-collected ForestPest dataset, the method achieves state-of-the-art performance, notably improving AP for small objects (AP_small = 0.131) and overall mAP (0.508), while maintaining practical compute costs. The work offers substantial advances for real-world forest monitoring and pest management, with implications for scalable, accurate remote-sensing-based ecological surveillance.

Abstract

Detecting agricultural pests in complex forestry environments using remote sensing imagery is fundamental for ecological preservation, yet it is severely hampered by practical challenges. Targets are often minuscule, heavily occluded, and visually similar to the cluttered background, causing conventional object detection models to falter due to the loss of fine-grained features and an inability to handle extreme data imbalance. To overcome these obstacles, this paper introduces Forestpest-YOLO, a detection framework meticulously optimized for the nuances of forestry remote sensing. Building upon the YOLOv8 architecture, our framework introduces a synergistic trio of innovations. We first integrate a lossless downsampling module, SPD-Conv, to ensure that critical high-resolution details of small targets are preserved throughout the network. This is complemented by a novel cross-stage feature fusion block, CSPOK, which dynamically enhances multi-scale feature representation while suppressing background noise. Finally, we employ VarifocalLoss to refine the training objective, compelling the model to focus on high-quality and hard-to-classify samples. Extensive experiments on our challenging, self-constructed ForestPest dataset demonstrate that Forestpest-YOLO achieves state-of-the-art performance, showing marked improvements in detecting small, occluded pests and significantly outperforming established baseline models.

Forestpest-YOLO: A High-Performance Detection Framework for Small Forestry Pests

TL;DR

The paper tackles detecting extremely small forestry pests in complex UAV imagery, where targets are highly occluded and data are imbalanced. It introduces Forestpest-YOLO, a YOLOv8-based detector augmented with SPD-Conv for lossless downsampling, CSPOK for robust cross-scale fusion, and VarifocalLoss to focus learning on hard/high-quality samples. On the self-collected ForestPest dataset, the method achieves state-of-the-art performance, notably improving AP for small objects (AP_small = 0.131) and overall mAP (0.508), while maintaining practical compute costs. The work offers substantial advances for real-world forest monitoring and pest management, with implications for scalable, accurate remote-sensing-based ecological surveillance.

Abstract

Detecting agricultural pests in complex forestry environments using remote sensing imagery is fundamental for ecological preservation, yet it is severely hampered by practical challenges. Targets are often minuscule, heavily occluded, and visually similar to the cluttered background, causing conventional object detection models to falter due to the loss of fine-grained features and an inability to handle extreme data imbalance. To overcome these obstacles, this paper introduces Forestpest-YOLO, a detection framework meticulously optimized for the nuances of forestry remote sensing. Building upon the YOLOv8 architecture, our framework introduces a synergistic trio of innovations. We first integrate a lossless downsampling module, SPD-Conv, to ensure that critical high-resolution details of small targets are preserved throughout the network. This is complemented by a novel cross-stage feature fusion block, CSPOK, which dynamically enhances multi-scale feature representation while suppressing background noise. Finally, we employ VarifocalLoss to refine the training objective, compelling the model to focus on high-quality and hard-to-classify samples. Extensive experiments on our challenging, self-constructed ForestPest dataset demonstrate that Forestpest-YOLO achieves state-of-the-art performance, showing marked improvements in detecting small, occluded pests and significantly outperforming established baseline models.

Paper Structure

This paper contains 19 sections, 7 equations, 3 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Advances in Small Object Detection
  4. Feature Fusion and Attention Mechanisms
  5. Loss Functions and Training Strategies for Robust Training
  6. Methodology: The Forestpest-YOLO Framework
  7. Overall Architecture
  8. SPD-Conv: Lossless Downsampling Module
  9. CSPOK: Cross-Stage Parallel Omni-Kernel Fusion
  10. VarifocalLoss (VFL) and Matching Optimization
  11. Experiments
  12. Dataset: The ForestPest Dataset
  13. Implementation Details
  14. Comparison with State-of-the-Art Models
  15. Ablation Study
  16. ...and 4 more sections

Figures (3)

  • Figure 1: The overall network architecture of Forestpest-YOLO, illustrating the main data flow from input to detection. Key modifications include the integration of the SPD-Conv module in the backbone and the CSPOK fusion block in the neck.
  • Figure 2: Detailed structure of the improved neck, highlighting the integration of our proposed modules for enhanced feature fusion. The diagram shows SPD-Conv for lossless downsampling and the CSPOK block replacing the original C2f unit.
  • Figure 3: Conceptual visualization of matching quality improvement achieved by using VarifocalLoss. The left side depicts potential poor matching, while the right (green lines) illustrates the precise target association our loss encourages.