Design And Implementation of An Embedded CNN based Weed Detection and Mechanical Weed Elimination Rover

Abstract: Weed invasion remains a significant challenge in modern agriculture, contributing to reduced crop productivity and increased operational costs. Conventional weed control practices, including manual weeding and chemical herbicide application, are labour-intensive and raise environmental and health concerns. This study presents the development, implementation, and experimental assessment of an embedded deep learning system based on computer vision for identifying weeds and removing them mechanically using a mobile robotic platform.

The proposed system employs a Convolutional Neural Network (CNN) deployed implemented on a compact embedded device like the Raspberry Pi to perform on-device weed detection from camera-acquired field images. Detected weeds are localized and processed in real time to guide a servo-driven mechanical plucking mechanism, enabling physical weed removal without the use of chemical herbicides. The system integrates visual perception, embedded processing, rover mobility, and mechanical actuation into a compact and low-cost platform intended for small- and medium-scale agricultural settings.

Experimental evaluation was conducted under controlled and limited real-field conditions to assess detection performance and system feasibility. The outcome indicate that the developed system can successfully identify weeds and perform targeted mechanical removal while operating within the computational constraints of embedded hardware. These findings demonstrate the capability of embedded deep learning robotic systems to support precise and chemical-free weed control solutions.

Keywords: Weed Detection, Convolutional Neural Networks, Precision Agriculture, Agricultural Robotics, Autonomous Rover, Edge AI, Mechanical Weed Removal.