Pine Nematode-Infected Wood Detection in Mixed Forests with YOLOv5s
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for identifying pine wood nematode-infected discolored woods in mixed coniferous and broadleaf forests suffer from low accuracy and difficulty in deploying deep learning models with numerous parameters on edge devices like drones, due to the complexity of forest environments and mixed tree species.
Innovation Solution
A modified YOLOv5s model is developed with a feature-filtering module, feature-enhancing module, and convolution-transformer module based on multi-head self-attention (MSHA) to enhance identification accuracy, and a multi-scale feature fusion layer to reduce computational complexity, allowing deployment on edge devices.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If deep learning algorithms are improved to enhance identification accuracy, then the accuracy of identifying pine wood nematode-infected discolored woods is improved, but the number of model parameters increases, making it difficult to deploy on edge-end devices
Solution Approach 1:
The model is divided into multiple functional modules: backbone feature extraction module, neck feature fusion module, and detection head module. Each module has specific functions and can be independently optimized, allowing accuracy improvement without proportionally increasing overall complexity
Solution Approach 2:
Different modules use different computational strategies tailored to their specific functions. The backbone uses standard convolutions for feature extraction, the neck uses coordinate attention mechanism for spatial feature enhancement, and the detection head uses simplified processing for bounding box prediction, optimizing the balance between accuracy and complexity in each local region
2Device complexity
If traditional manual inspection is used, then the identification process is simple, but the workload is huge, time period is long, and it is difficult to comprehensively grasp spatial distribution of infected trees in large areas
Solution Approach 1:
Manual inspection is replaced with an automated deep learning-based image recognition system that processes drone-captured forest images. This substitution transforms the mechanical manual examination process into an automated computational system, dramatically improving identification efficiency and enabling comprehensive coverage of large forest areas
Solution Approach 2:
The system creates digital copies (images) of forest areas through drone photography and processes these copies through the deep learning model. This allows comprehensive analysis of spatial distribution of infected trees without physically examining each tree, significantly improving productivity while maintaining simplicity
3Measurement precision
If deep learning models with large number of parameters are used, then the identification accuracy is improved, but the deployment on edge-end devices such as drones becomes difficult
Solution Approach 1:
The model architecture is segmented into modular components that can be independently deployed or optimized. The backbone, neck, and detection head can be separately trained and combined, allowing flexible deployment on edge devices with limited resources while maintaining high accuracy through coordinated optimization of each segment
Solution Approach 2:
The model uses parameter sharing and weight freezing techniques where certain layers are trained with fixed parameters while others are optimized. This reduces the number of trainable parameters that need to be stored and processed on edge devices, making deployment easier while preserving identification accuracy through selective parameter optimization
4Device complexity
If manual inspection is used in mixed forests, then the method is simple, but the accuracy of identifying pine wood nematode-infected discolored woods decreases due to increased image complexity
Solution Approach 1:
The coordinate attention mechanism in the neck module serves as an intermediary that processes and enhances spatial features from the backbone before passing them to the detection head. This intermediary mechanism specifically addresses the complexity of mixed forest images by enhancing spatial relationships and feature discrimination, improving identification accuracy without requiring overly complex detection methods
Solution Approach 2:
The model combines multiple types of convolutional operations (standard convolutions, depthwise separable convolutions, and coordinate attention mechanisms) to create a composite architecture that effectively handles the complexity of mixed forest images. This composite approach leverages the strengths of different computational methods to improve identification accuracy in complex environments
Data Source
AI summary
The present disclosure provides a method for identifying a pine wood nematode-infected discolored wood in a mixed coniferous and broadleaf forest, including inputting a pine forest image of the mixed coniferous and broadleaf forest to be identified into a trained identification model for the pine wood nematode-infected discolored wood to identify the pine wood nematode-infected discolored wood in the mixed coniferous and broadleaf forest, wherein the trained identification model is improved based on a you Only Look Once version 5 small (YOLOv5s) model by connecting a feature-filtering module after a Neck; constructing a feature-enhancing module to replace a C3 module in a Backbone; constructing a convolution-transformer module based on multi-head self-attention (MSHA) to connect after a last layer of residual units in the Backbone; constructing, by Group Shuffle Convolution (GSConv), a multi-scale feature fusion layer to replace an ordinary convolution in the Neck.