A Method of Object Detection Based on Feature Fusion

Abstract

The invention discloses a target detection method based on feature fusion, which aims to solve the shortcoming of low detection accuracy of small-sized targets in the current detection method. The technical solution is to construct a target detection system composed of a feature extraction module, an indirect feature fusion module, a direct feature fusion module, two deformation modules, and a detection module; the trained target detection system is used to perform feature extraction, indirect feature fusion, direct Feature fusion to detect and identify the location and category of the target. Among them, the indirect feature fusion module splices the high-level and low-level feature maps into one feature map, then calculates the dependencies between different pixels for the spliced ​​feature maps, and shares the dependencies between different feature maps, realizing the integration of high-level features and low-level features. The direct feature fusion module transfers the high-level feature map information to the low-level feature map layer by layer, which enhances the semantic and location information in the low-level features of small-sized objects and improves the detection accuracy of small objects.

Description

technical field [0001] The invention relates to the technical field of computer vision, in particular to a method for object detection based on feature fusion. Background technique [0002] Target detection is one of the important research directions in the field of computer vision. The traditional target detection method is to extract features by constructing feature descriptors (such as histograms of orientation gradients, etc.) for images in a certain area, and then use classifiers to classify the features. Target detection, such as support vector machine SVM (Support Vector Machine), etc. Recently, with the development of convolutional neural networks, engineering features have mostly been replaced by convolutional neural network features, and object detection systems have made great progress in both accuracy and speed. [0003] Currently, object detection methods based on deep learning are divided into two-stage detection methods and single-stage detection methods. ...

AI Technical Summary

Problems solved by technology

[0007] The technical problem to be solved by the present invention is to solve the shortcoming of low detection accuracy of small-sized targets in the current single-stage detection method

Images