Advancing Weed Management: A Novel Image Classification System Using NCA-based Feature Selection and PSO-GSA Optimized Random Forest Classifier

Abstract

Adopting more intelligent and targeted weed management strategies enables farmers to reduce their dependence on chemical treatments, save valuable time and resources, and improve crop yields, all while lessening negative effects on the environment and human health. To address these urgent issues, it is essential to pursue a more sophisticated agricultural approach that incorporates automated technologies and leverages machine learning algorithms. This study presents an innovative weed detection system aimed at differentiating crops from weeds by integrating various feature extraction techniques with advanced machine learning capabilities. At the core of this system is the application of hybrid features, coupled with an effective feature selection method based on neighborhood component analysis. These distinctive features are utilized by a particle swarm optimization and gravitational search algorithm (PSO-GSA) optimized random forest classifier, which effectively categorizes images into either crops or weeds. The results demonstrate that our method, which combines hybrid feature extraction with the PSO-GSA-RF classification approach, significantly outperforms other techniques. Furthermore, this system can be integrated into agricultural robots, allowing for the precise application of herbicides only where necessary, thereby minimizing the introduction of harmful chemicals into the food supply and reducing the risk of human exposure to dangerous substances.