Tiny target detection method based on pixel-level enhanced feature multi-scale fusion

Abstract

The invention provides a tiny target detection method based on pixel-level enhanced feature multi-scale fusion, and the method comprises the steps: firstly carrying out the labeling of each pixel point of an original input image and the labeling of a whole image, and then carrying out the feature extraction and enhancement of the original input image; multi-scale feature layers are extracted and fused, weight distribution is carried out on the fused feature layers, and finally the category of the tiny target is obtained. According to the method, the tiny target detection task can be well met, the method is efficient and accurate, the precision and efficiency of tiny target detection are improved, the method has high application value and economic benefits, and it is proved through actual verification that the method can be well applied to the tiny target detection task on the surface of an object.

Description

technical field [0001] The present invention relates to the technical field of object detection on the surface of any object, in particular to the detection of tiny objects on the surface of the object. Background technique [0002] The detection of tiny targets is used in many fields, including machinery manufacturing, medicine, construction, electronics and other fields. However, the task of detecting tiny objects on the surface of objects has always been a big challenge in terms of accuracy, and its low-precision detection results are far from meeting the detection requirements. [0003] With the improvement of modern intelligence level, people have more and more stringent requirements for target detection in machine vision, not only high accuracy and efficiency, but also smaller and smaller targets. The traditional small target detection adopts manual detection method. There is no unified standard for manual detection, and the rate of missed detection and false detectio...

Images