DeepWeed: A Dynamically Adaptive Lightweight Deep Learning Model for Efficient Weed Classification

This work introduces DeepWeed, a lightweight deep-learning model inspired by the VGG16 architecture for image-based weed classification. Leveraging transfer learning and L2 regularization, DeepWeed aims to achieve high accuracy while maintaining computational efficiency. Our study evaluated the model’s performance on a dataset of five weed species. While an initial accuracy of 73.6% was achieved without data augmentation and L2 regularization, incorporating these techniques significantly improved, reaching an accuracy of 92.4 %. This increase highlights the effectiveness of data augmentation and L2 regularization in enhancing model performance. DeepWeed’s combination of high accuracy and computational efficiency suggests its potential as a practical tool for weed classification in agricultural applications, potentially contributing to improved farm management and reduced environmental impact.