Multi-specialized target detection algorithm based on deep learning

Abstract

The invention discloses a multi-specialized target detection algorithm based on deep learning. A detection model framework of the target detection algorithm mainly comprises a RoI generation module and a multi-specialized channel module, wherein the target detection algorithm obtains an augmented RoI set through a multi-scale sliding window and selective search, and takes over the multi-scale sliding window in an exhaustion mode to generate a dense RoI set; a plurality of special channel networks are designed, each special channel network is responsible for detecting one type of RoI, and eachspecial network is provided with a full connection layer in the same form. According to the method, two strategies are used for improving the performance, the two strategies enhance the performance ofthe detection model in different aspects, and multiple special channel detection related to the shape category is designed; an augmented RoI dataset is further generated using a selective search andprovided to the network for training.

Description

technical field [0001] The invention relates to a target detection algorithm, in particular to a multi-specialty target detection algorithm based on deep learning. Background technique [0002] At present, object detection, as one of the classic research contents in computer vision, has received more and more attention in the research field. In general, object detection uses unique shape patterns as evidence to find objects of interest in images. Object detection models are trained on these shape patterns that exhibit classes of different objects. However, it is quite cumbersome for a single model to accurately recognize all the appearances, since objects appear quite differently in images according to their basic shape as well as different poses and viewpoints. Therefore, traditional object detection methods often adopt the method of mixing multiple classifiers, and each classifier is only associated with the corresponding shape pattern in order to better capture the shap...

AI Technical Summary

Problems solved by technology

However, it is rather cumbersome for a single model to accurately identify all appearances, since objects appear to vary greatly in images depending on their basic shape as well as different poses and viewing angles.

Images