A Knife Switch Status Recognition Method Based on Improved Deep Learning

Abstract

The invention discloses a knife gate state recognition method based on improved deep learning, comprising the following steps: obtaining a training model; predicting an input image through the training model to obtain the probability of a prediction frame; using a sliding window strategy to select a candidate area and obtain a label; The candidate area is deleted to obtain the candidate rectangular frame; the candidate rectangular frame is fitted with a straight line to obtain an accurate rectangular frame; the state of the knife switch and insulator in the precise rectangular frame is judged to complete the identification of the knife switch state. The present invention adopts the convolutional neural network improved by the pooling strategy based on space weighting to obtain the training model on the image set; then the potential position of the insulator and the switch is detected through the training model, and a variety of switches are identified according to the connectivity with the insulator The closed or disconnected state of the switch can accurately locate the position of the insulator and the switch, and significantly improve the accuracy of the status recognition of the switch.

Description

technical field [0001] The invention relates to the field of pattern recognition and classification, especially the classification and detection of specific targets, and in particular to a method for recognizing the state of a knife switch based on improved deep learning. Background technique [0002] Substation real-time monitoring technology has developed rapidly in recent years, among which the power equipment detection and recognition technology based on image processing has become a research hotspot. Since the images captured in reality often contain many other targets rather than the power equipment target itself of interest, and the background of the captured images is also more complex, such as different lighting conditions, shooting angles, etc., which makes the same target in different Different patterns are present in the image. [0003] Traditional identification methods for power equipment (such as insulators and knife switches) mainly rely on target color feat...

AI Technical Summary

Problems solved by technology

However, these methods all have the problem of mutual interference between multi-knife gate targets, and the recognition accuracy still needs to be further improved.

Images