A real-time detection method for a single-stage multi-scale specific target based on an effective receptive field

Abstract

The invention provides a real-time detection method for a single-stage multi-scale specific target based on an effective receptive field. The method comprises the following steps: firstly, extractinga corresponding feature layer from a multi-scale architecture of an SSD, and selecting a scale according to a pixel range covered by a receptive field; secondly, an anchor structure in a traditional method is removed, fewer feature layers are adopted, and classification and regression are directly carried out on corresponding receptive field frames of the feature map by utilizing the characteristics of a natural receptive field; Finally, an RF (receptive field) sampling frame ash placement learning strategy is adopted, and redundant parameters are prevented from being learned. According to themethod, the complexity of a traditional algorithm based on an anchor sampling frame is greatly reduced, the detection efficiency is improved, the real-time effect can be achieved, and the method hasvery high use value under the application background with very large data volume.

Description

technical field [0001] The invention belongs to the field of computers and relates to a single-stage multi-scale real-time detection method for a specific target based on an effective receptive field. Background technique [0002] Big data can be considered as a hot spot in the current academic and industrial research and gradually affects people's daily life and work style. Its characteristics can be considered as large amount of data and diversity, so real-time becomes very important. [0003] With the development of computer and artificial intelligence technology in the image field, target detection has become a very hot topic, and specific target detection is a key step in many subsequent specific target-related applications, such as: specific target recognition, specific target verification, and specific target tracking etc. Convolutional neural networks have achieved remarkable success in recent years. From image classification to object detection, it also motivates...

AI Technical Summary

Problems solved by technology

[0009] Although the anchor-based method has achieved high accuracy and efficiency in today's target detection tasks, however, due to the complexity of its algorithm (a*a*B, a*a is the size of the feature map, B is the number of anchors )

As the amount of data continues to increase, the efficiency of the detector will drop sharply, for example: the calculation efficiency of Faster-RCNN is very low

Images