An Apple Detection Approach Based on Deep Learning Algorithm in Kashmir Orchards for Apple Harvester Robot

摘要

The apple industry in Kashmir, India, is one of the cornerstones of the region's economy. Efficient apple harvesting is crucial for both productivity and fruit quality. This paper presents an innovative approach for apple detection in Kashmir orchards using a deep learning algorithm, YOLOv8, integrated into a RealSense camera and an Apple Harvester Robot. The proposed method aims to enhance the efficiency and accuracy of the apple harvesting process, addressing the unique challenges posed by the varying lighting conditions, foliage, and diverse apple shapes and colors. YOLOv8, a state-of-the-art object detection model, is employed to detect apples in real-time. We fine-tuned the YOLOv8 model on a comprehensive dataset of annotated apple images from Kashmir orchards, ensuring that the model can effectively identify apples in different stages of maturity and amidst varying environmental conditions. Our approach not only detects apples but also localizes them with bounding boxes, providing precise information for the robot's manipulator to harvest the fruit with minimal damage. The algorithm we proposed has achieved an average of 91.2%. The recall and precision rates were 98% and 93.9% respectively. All the classifiers correctly predicted instances, in the dataset.