Apple recognition based on Convolutional Neural Network Framework

Robots are important tools for agricultural modernization. Fruit picking robot has become a research hotspot. How to extract the fruit recognition accuracy has become the key technology. In this paper, an advanced target detection framework, single shot multi-box detector (SSD), is used to detect apples in the orchard. The environment of the apple orchard is complex, and there is often occlusion between fruits. So it is a difficult task to identify apple. SSD algorithm uses Convolution Neural Network to extract the apple characteristics of orchard automatically, This method has a higher recognition accuracy than the traditional artificial feature recognition method, which meets the real-time requirements. Previous work used Faster R-CNN to detect apple and achieved good results. In the SSD network, the convolution layer and the full connection layer are converted into the complete convolution layer, which improves the speed of fruit detection.