Image target detection method based on novel feature pyramid network

Abstract

The invention discloses a novel feature pyramid network-based image target detection method, which comprises the following steps of: firstly, processing an input highest-level feature map by adopting a pyramid pooling module to obtain multi-scale features containing context information of regions with different sizes of the feature map, and then carrying out dimensionality reduction and average pooling on all levels of feature maps to obtain a multi-scale feature map; the method comprises the steps of obtaining four feature maps with the same channel number and size, sending the four feature maps into two 1 * 1 convolution layers to calculate and obtain position weights, performing position weighting on the four feature maps, processing the feature map with the highest level by adopting 3 * 3 convolution, adding and fusing the feature maps from top to bottom to obtain five feature maps, and finally predicting on the five feature maps. And obtaining a target detection result. According to the method, the problems of aliasing effect, dilution of high-level semantic information and limited access of low-level spatial information can be solved, the fusion structure of the feature pyramid network can be not complicated, features with higher discrimination can be generated, and the detection precision of different types of detectors can be improved.

Description

technical field [0001] The invention relates to the technical field of digital image processing, in particular to an image target detection method based on a novel feature pyramid network. Background technique [0002] Target detection refers to the estimation of the position and category of objects in the image at the same time. It can be used in modern society such as automatic driving, helmet detection, pedestrian detection and other scenarios. The current mainstream target detection algorithm is suitable for different scale targets, especially small Target detection is poor. How to solve this problem and obtain more accurate detection results has become a current research hotspot. [0003] The current mainstream object detection algorithms can be roughly divided into two-stage detectors and single-stage detectors. The two-stage object detector first generates some predefined, dense sliding windows on the feature map through the proposal mechanism to determine the most ...

AI Technical Summary

Problems solved by technology

[0006] The purpose of the present invention is to overcome the deficiencies of the prior art and provide a new image target detection method based on a feature pyramid network, which can not only effectively solve the aliasing effect caused by the fusion structure of the traditional feature pyramid network, the dilution of high-level semantic information and the low-level The problem of limited access to spatial information can also not complicate the fusion structure of the feature pyramid network, generate more discriminative features, and improve the detection accuracy of different types of target detectors

Images