Weedbot: Weed Easy Extraction Using Deep Learning with a Robotic System

Abstract

Weeds refer to plants that a farmer does not intentionally cultivate because they interfere with or limit the growth and yield of crops. Due to their fast reproduction and spreading, weeds are well-known for quickly increasing in population over large areas. From the very beginning of the agricultural revolution, weed control used to be done by hand but as time and technology advanced so did methods for managing unwanted plants. Human Intervention or using a single-row weed cutter, weeds are removed from fields. Due to the difficulties of eliminating the background class from the image and increasing the accuracy of the navigation system of the bot, as well as the fact that it does not support the technique of extracting weeds from paddy fields, a vehicle method for removing weeds from the field has not been created. These difficulties have been tackled by combining computer vision with automation, minimizing the need for human exertion. The project aims to create a robotic chopper that deal with weeds and employ CNN (Convolutional Neural Network) algorithm for image categorization of weeds and crops. We use image processing techniques in the proposed system to identify weeds. The system is trained with 590 images of weeds and crops, and testing is performed with 254 images. The system obtained an accuracy of 80% towards processing weeds and then chopped by the robotic arm. The robotic arm cutter that has great precision and is simple to operate, inexpensive and does not require hazardous chemicals.