Target detection method and system based on lightweight deformable convolution

Abstract

The invention discloses a target detection method and system based on lightweight deformable convolution. The method comprises the following steps: a deep separable feature extraction network is created as a source network; a target network is created by using an algorithm structure which is the same as that of the source network but not performing deformable convolution layer replacement; distance loss approximation I sperformed on the output of the source network and the target network by simulating features extracted by multiple network layers of the source network; before the classification layer, the last layer of feature extraction enables the source network to approach the output of the target network layer; a new feature extraction network model is acquired after joint training through a mutual learning framework of the source network and the target network; and the new feature extraction network model is taken as a feature extractor to perform feature extraction of the image data to complete target detection. Through mutual learning of the target network and the source network, the identification precision is improved, the data operation amount is reduced, and the burden on hardware equipment is reduced.

Description

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine

Login to View More

AI Technical Summary

Benefits of technology

This technology allows for better understanding by converting important factors like image quality into values that help train neural networks or predict future performance more accurately than previous methods. It also makes it easier to learn new models quickly while reducing computational load compared to existing techniques.

Problems solved by technology

This patented technical solution described in the current text discusses how difficultly detecting specific things like people or cars within images captured under varying environmental factors (lighting) leads to difficulty in accurately identifying them. Different methods have been developed over time, including convolutional neural networks trained specifically against these environments, fuzzy logic systems, decision trees, etc., but they all face similar issues when trying to identify unknown items at low resolutions.

Images