Weed Detection in Farmlands Using RCNN

Abstract

Detection and classification of weeds are important for the optimization of farming applications as it allows to identify species so that they can be removed, competing with others who need nutrients to grow their crop. A variety of methods have been used in different research studies but there still exists a research gap for a comprehensive comparative analysis of multiple machine learning models. Here, we attempt to provide a comparative study for classic/not ensemble (Support Vector Machines (SVM), Random Forest etc.) machine learning models with Region-Based CNNs (RCNN). Last, we stack LSTM (a type of RNN) onto the model to making it more powerful with respect to time learning and performance.1300 images of sesame crops and different types of weeds, serving as a proxy for diverse agricultural scenarios. The proposed weed detection and classification technique demonstrates strong potential for practical applications, particularly when integrated with autonomous robots for real-time field deployment. The study contributes to the field of precision agriculture by providing comparative insights into machine learning and deep learning models, emphasizing the importance of temporal learning for weed management.