Robotic Harvesting Systems with Real-Time Fruit Ripeness Detection

Abstract

The growing concerns on automated agricultural applications have brought into light on the importance of effective robotic harvesting where selective operation needs in fruit picking are to be agreed on. The following paper proposes the concept of an intelligent robotic harvesting system combined with real-time fruit ripeness detection to minimize food waste and maximize the quality of harvest. The system suggested would be a multispectral computer vision system based on deep learning algorithms to evaluate the ripeness parameters (e.g., color, texture, and sugar content) with an accuracy rate of 92. 03, compared to the traditional methods of computer vision, using RGB, is 0.75 higher. A custom soft-gripper robotic manipulator with force feedback allows the fruits to be handled gently, even with an 85% successful picking rate at 5 seconds per fruit on-field tests. Notable advances can be summarized by two points: (1) a lightweight CNN-Transformer hybrid network to be applied to edge devices, and (2) a path planning algorithm with reduced collision likelihood in dense foliage. As per the experimental outcomes on the strawberry and tomato crops, a 30 percent increase in the yield retention is achieved over manual harvesting. This system efficiently fills in the existing gaps in the agricultural robotics field, where speed and precision should be balanced with flexibility to adapt to different orchard conditions, leading to scalable autonomous agriculture.