Semantic segmentation method and system for bolt in power transmission line

Abstract

The invention discloses a semantic segmentation method and system for a bolt in a power transmission line. The method comprises the following steps: positioning a region where a bolt is located in an obtained to-be-recognized image of a power transmission line so as to acquire a positioning image of the region where the bolt is located; separately acquiring a low-level feature map with detail features and a high-level feature map with semantic information of the positioning image; fusing the low-level feature map and the high-level feature map; and acquiring a pixel-level semantic segmentation result of the positioning image based on the fused feature map so as to realize defect classification of the bolt in the power transmission line. The method has the advantages that most of the background of the image is filtered out through the bolt area positioning image, background interference is effectively reduced, and bolt recognition precision is improved; and by fusing multi-scale cavity convolution pooling features and multi-dimensional high-low layer feature information, the classification refinement degree of the bolt is improved, and semantic segmentation of the bolts of the power transmission line is realized.

Description

technical field [0001] The invention relates to the technical field of digital image processing, in particular to a method and system for semantically segmenting power transmission line bolts. Background technique [0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art. [0003] With the rapid development of deep learning technology, object detection algorithms based on neural networks have been widely used. The neural network realizes the classification and recognition of the target by self-learning the characteristics of the target. [0004] There are many components in the transmission line. For large objects such as clamps, anti-vibration hammers, insulators, etc., due to the large proportion of the image in the image, the characteristics are more obvious, and the neural network can easily learn the characteristics of large-scale objects during training. Significant featu...

AI Technical Summary

Problems solved by technology

[0006] For a small-sized target such as a bolt, the area in the image is relatively small, and the feature information such as brightness, chromaticity, and edge is relatively small, and the size of the bolt and the background is not balanced. If the neuron uses a small receptive field, it is difficult to capture the overall situation. information, with a large receptive field, it is easy to lose the characteristics of the bolts. Therefore, it is difficult to perceive the characteristic information of the bolts during neural network learning, resulting in poor recognition results; in addition, the inventors found that the defect recognition of bolts belongs to fine-grained images Recognition problems, fine-grained targets are more difficult to identify because most of the features are similar. For example, a bolt with a pin and a bolt without a pin. The difference between the two is only on the small pin. When the pair of bolts is missing When identifying this defect, it is easy to misjudge a bolt with a pin as a bolt without a pin

Therefore, for the defect identification of bolts, it is difficult to achieve the desired effect only by relying on the traditional deep learning target detection algorithm. It is necessary to consider how to enrich the characteristics of bolts.

Images