Object recognition methods, devices and single-step object recognition neural networks

摘要

PROBLEM TO BE SOLVED: To provide an object recognition method, an apparatus, and a single-step object recognition neural network. By using extended convolution instead of part of the convolution layer in a conventional single-shot multi-box detection (SSD) network, object recognition is achieved by reducing the size limitation of the input image and increasing the connection path. By connecting the convolution layers using a residual structure, the convergence effect is ensured and the accuracy of object recognition is improved as a whole. [Selection diagram] Fig. 3