Computer Vision Assisted Real-Time Bird Eye Chili Classification Using YOLO V5 Framework

Abstract

Computer vision-based classification systems have become increasingly popular in the agricultural industry in recent years. This paper proposes a computer vision-assisted bird eye chili or 'kantahri mulaku' classification framework using the You Only Look Once V5 (YOLO V5) object detection model. Automated sorting systems based on computer vision can accurately identify and classify chilies based on attributes such as size, shape, colour, and texture. The dataset for the research consists of images of bird-eye chilies in different positions and backgrounds. The model was trained using this dataset, and it could correctly identify and categorize bird-eye chili. The chilies was then picked up by a robot manipulator and sorted by ripeness. Bird-eye chili images captured in real agricultural situations have used to assess the effectiveness of the suggested framework. Images of red and green chili was taken from above using a high-resolution Raspberry pi 4B camera attached to a custom-built 3-degrees-of-freedom (DoF) robot arm. We used public and real-time images to train the YOLO algorithm on photographs of bird-eye chili captured in real-time. As the robot arm goes around the chili plants, this model is connected with the robot's software control system to allow real-time detection and localization of the chili's. By automating bird-eye chili crop monitoring and management, this system has the potential to significantly contribute to the growth and viability of the agricultural sector. We got a mAP of 0.94 and an average accuracy of 0.90 with the suggested method. Using a robotic manipulator for chili grading improves productivity and reduces human error compared to traditional methods. To test the robustness of the YOLO V5 framework, it has implemented on the Raspberry pi 4B graphical processing unit (GPU) computer.