WeedLiteNet-V2: A Lightweight Residual Multi-Scale CNN with Efficient Channel Attention for Weed Species Classification
Contributors
Aayush Shrivastava
Dr Ajay Kumar
Keywords
Proceeding
Track
General Track
License
Copyright (c) 2026 Sustainable Global Societies Initiative

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.
Abstract
The present study reports on the development of WeedLiteNet-V2, a light-weight nine-class weed classifier for DeepWeeds. The model uses depthwise-separable convolution, residual learning, parallel 3x3 and 5x5 depthwise branches, and Efficient Channel Attention (ECA). According to project documentation, the model has 373,839 parameters, achieved a best validation accuracy of 94.97%, and a test accuracy of 94.50% (2,490/2,635). However, these results are preliminary since there was inconsistency in the split count provided, and neither the final checkpoint nor the predicted classes were made available for independent verification.