Visual Recognition and Location Algorithm of Strawberries Based on Enhanced YOLOv7 Detector and RGB Depth Camera

Currently, the primary technology obstacle to the application of robots for strawberry picking is the meagre rate of harvest. Consequently, in order to determine the strawberry stem's picking point, Accuracy and real-time functionality of localisation techniques need to be enhanced. The accuracy of the localization algorithm can be increased by estimating the posture of the fruit axis's direction, or the fruit goal. Research suggests a new strawberry harvesting robot for ridge-planted plants in addition to a fruit pose estimator known as rotated Yolo (Enhanced YOLOv7), a (YOLOv7) object detection algorithm, and Red Green Blue-Depth (RGB-D) detecting. It greatly improves the picking point’ accuracy of localization. The YOLOv7 model has been altered to include more effective modules, attention processes, the removal of unnecessary feature layers, and layers specifically for the identification of tiny targets. The result of these changes is an enhanced and lightweight YOLOv7 network model. The operation speed was significantly enhanced by the streamlined network architecture. Secondly, the training set's labels and anchors were set using the rotating angle variable. Then, using the rotating anchors, and logistic regression was used to forecast how the box boundaries of the target fruits would spin. Based on 100 strawberry picture tests, the average identification and recall rates for the suggested model were 94.43% and 93.46%, respectively.