Application of back propagation neural networks on maize plant detection

Abstract

Development of automated in-row weed control is one of the costliest and complicated tasks in agricultural industry despite the rapid development of agricultural robotics. Hence, this study proposes an easy-to-implement and accurate system capable of real-time maize plant detection, which is the key part of the entire weeding machine. Mediterranean farmers use mechanized equipment for dominant crops, however, they suffer a labor-intensive in-row hand weeding. Therefore, this work focuses on a Back propagation neural network system to be a framework for a real-time maize plant classifier utilizing advanced machine-vision (single-lens vision) techniques. Back Propagation Neural Network (BPNN) incorporates a single-board computer platform. The proposed framework is tested on images that on images that have no-specific distinguishing geometric pattern, varying light conditions. The obtained BPNN results were found to be encouraging considering the time consuming occurs manually to differentiate the maize plant from the other harmful herbs.