Crop-Weed Detection, Depth Estimation and Disease Diagnosis Using YOLO and Darknet for Agribot: A Precision Farming Robot

Abstract

Precision agriculture depends on technology and data to maximise crop yields and lower production costs. Precise identification of weeds in crop fields reduces the risk of crop damage and optimises herbicide use. For this, a YOLO model is used, which is a deep learning-based real-time object detection approach employing convolutional neural networks (CNNs). It requires a single forward pass through a neural network, making it computationally efficient. Estimating the distance of each weed/crop plant from the camera is achieved through a monocular approach, using reference images and the camera’s focal length. The distance, found to be accurate to a sufficient degree, can be used for targeted herbicide application. A flask application, trained on a data set of healthy and sick leaves, is created for identifying crop diseases. It can determine the type of disease present, which enables early disease detection and treatment. Agribot, an autonomous robot, integrates these technologies and performs tasks like spraying fertilisers and water and uprooting weeds using its delta arm mechanism, making it an effective solution for precision agriculture.