Image Based Rice Weed Identification Using Deep Learning and Attention Mechanisms

Weed management is a critical aspect of modern agriculture, directly impacting crop yield and quality. In rice cultivation, weeds can significantly reduce productivity, leading to economic losses and increased environmental impact due to the excessive use of herbicides. In this study, authors have proposed a model ResNet50 based on convolutional neural networks (CNNs) architecture and attention mechanism module to develop a robust and efficient system for weed detection and classification. A comprehensive dataset of images is also developed, containing two major weed species, Cyperus difformis, and Echinochloa colona, commonly found in rice fields and healthy rice plants, to train and validate our deep learning models. The proposed system is designed to detect the presence of weeds in real time by analyzing images captured within the rice fields. It can distinguish between weeds and healthy rice plants with a testing accuracy of 97.40% and an average F1 score of 96.5%. Additionally, our model can be deployed into mobile applications for weed detection for implementing targeted weed management strategies for farmers.