Multi-Growth Period Tomato Fruit Detection Using Improved Yolov5

Abstract: In agricultural mechanized production, in order to ensure the efficiency of hand-eye cooperative operation tomato picking robot, recognition accuracy and speed multi-growth period fruit is an important basis. Therefore, improve while ensuring or improving accuracy, this paper improves Yolov5s model by adding architecture lightweight mobilenetv3 model. Firstly, deep separable convolution replaced backbone network Yolov5s, which reduces amount operation. Secondly, linear bottleneck inverse residual structure fused obtain more features high-dimensional space perform low-dimensional space. Third, attention mechanism inserted into last layer highlight accuracy. The research results show that improved Yolov5 remains above 98%, CPU 0.88f·s-1 faster than GPU 90 frames per second Yolov5s. Finally, a set software system designed developed using RealSense D435i depth camera PYQT. further verifies feasibility model, lays foundation for visual design robot recognition.