A deep learning approach to weed eradication in precision agriculture

This study develops a deep learning framework for the precise identification of weeds in wheat fields to enable targeted herbicide application. The research evaluates and compares the performance of six convolutional neural network (CNN) architectures, including ResNet 50, ResNet 101, GoogLeNet, SqueezeNet, ShuffleNet, and DarkNet 19, under real world conditions. A dataset captured in Peshawar, Pakistan, across various growth stages and environmental conditions was used for assessment. Model robustness was enhanced through data augmentation and color space alterations. Performance was rigorously evaluated using precision, recall, F1 score, and Area Under the Curve metrics. SqueezeNet emerged as the most efficient architecture, achieving an optimal balance between high classification accuracy and low computational complexity, making it ideal for real time applications. ResNet 50 and GoogLeNet also demonstrated strong performance. The findings indicate that plant maturity significantly improves classification accuracy. This work conclusively shows that lightweight, efficient CNNs can form the core of a practical, on‐site precision farming system, promoting sustainable weed management.