Pin screw defect detection method and system based on semantic segmentation and spatial relationship

Abstract

The invention discloses a pin screw defect detection method and system based on semantic segmentation and a spatial relationship. The method comprises the following steps: acquiring a power transmission line picture; labeling the power transmission line picture and training by using a semantic segmentation network; segmenting a to-be-detected image by using the model, segmenting a set square and a screw region, and obtaining an initial recognition result; According to the spatial context relation between the pin screws and the set square, eliminating common screws which are wrongly recognized in the initial recognition result; drawing an external rectangle of the set square area; according to the circumscribed rectangles, constructing rectangular areas near three vertexes of the three set squares; finding out the central point of the pin screw, judging whether the central point is in the rectangular area, and if so, detecting whether the pin screw has defects; and if not, indicating that the picture does not have the screw with the pin. According to the invention, the detection capability of small targets is improved, so that the detection process is more accurate, reliable, rapid, efficient and more intelligent.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to a defect detection method and system for pinned screws based on semantic segmentation and spatial relationships. Background technique [0002] The problem of screw defect diagnosis on transmission lines is classified as a small target recognition problem, which has always been a major challenge in the field of computer vision research "Kang Lingzhou, Chen Fushen, Wang Desheng. Infrared image small based on morphological algorithm Research on Target Detection Method[J]. Optoelectronic Engineering, 2010,37(11):26-31". There are two main reasons for the difficulty in identifying small objects: one is that the size of the small object is too small, and the continuity of its internal features is weak, resulting in a very low dimensionality of the extracted features, and the classifier is prone to overfitting. "Chen Jiangyun .A new small target detection method based on deep ...

AI Technical Summary

Problems solved by technology

However, the experiment found that this method has limited effect on improving the segmentation of small objects, because only a single-layer non-learning upsampling method is used to convert the low-resolution feature map into a high-resolution feature map, and a lot of image information is still lost. " Hu Tai. Research on small target semantic segmentation algorithm based on deep neural network [D]. Nanjing: Nanjing Normal University, 2018”

Images