Implementing Innovative Weed Detection Techniques for Environmental Sustainability

Abstract

Agriculture, supporting over half of India's population, grapples with the challenge of weed control. Current methods applied in plantation crops lack efficiency and pose environmental and health risks. This paper advocates a paradigm shift, emphasizing the critical need for effective weed detection using cluttered unmanned aerial vehicle (UAV) images. The research methodology integrates image processing, Mask R-Convolutional Neural Networks (R-CNN), and Internet of Things (IoT). A dataset of 200 UAV images was subjected to a thorough preprocessing. In the initial phase, weeds and crops were identified with precision employing an UAV-tailored Mask R-CNN with instance segmentation. This was found to surpass traditional methods in terms of communication between the model and the agricultural environment. For timely decision-making, real-time data were collected using IoT. Average Precision (AP) values reveal high accuracy, notably 89.1% for weeds, 88.9% for crops, and an overall precision of 89.4%. The Mask R-CNN network segments and classifies images, marking weed zones communicated to farmers via Raspberry Pi with a GSM module, enabling real-time alerts and informed decision-making for efficient weed control. This holistic approach, providing object classifications, detailed bounding boxes, and masks, addresses weed control challenges, highlighting the transformative potential of advanced technologies in agriculture.