We introduce a grouped dropout strategy and modify the CNN architecture to improve the accuracy of multi-class insect recognition. Leveraging the Inception module’s branching structure and the adaptive grouping properties of the WeDIV clustering algorithm, we developed two grouped dropout models, the iGDnet-IP and GDnet-IP. Experimental results on a dataset containing 20 insect species (15 pests and five beneficial insects) with 73,635 images demonstrated an increase in cross-validation accuracy from 84.68% to 92.12%, with notable improvements in the recognition rates for difficult-to-classify species. Our model showed significant accuracy advantages over standard dropout methods on independent test sets, with much less training time compared to four conventional CNN models, highlighting the suitability for mobile applications.
Grouped dropout-based CNN for insect pest recognition. (A) Architecture of GDnet-IP; (B) Inception-based GDnet-IP, where the grey branch is randomly deactivated; (C) Clustering-based GDnet-IP, where the channels in 'Group 2' are randomly deactivated.
GDnet-IP has been tested with Python 3.?? and PyTorch 2.??. The user can easily set up the required environment using Conda by following these steps:
-
Clone the repository
git clone https://github.com/ZhijunBioinf/GDnet-IP.git cd GDnet-IP -
Create and activate Conda environment
conda env create -f environment.yml conda activate GDnet-IP
To get started with GDnet-IP, you can load the GDnet-IP models directly by using the following Python script:
from ?? import ??
# Load the GDnet-IP model
model = Implementing: Dongcheng Li (dongchengli287@gmail.com)
Supervisor: Zhijun Dai (daizhijun@hunau.edu.cn)
Dongcheng Li, Yongqi Xu, Zheming Yuan, Zhijun Dai*. GDnet-IP: Grouped Dropout-Based Convolutional Neural Network for Insect Pest Recognition. Agriculture, 2024, 14(11), 1915.