Research On Image Recognition Of Hydraulic Line Leakage

Hydraulic oil leakage will not only pollute the equipment and the environment,but also lead to equipment can not work normally,and even cause safety accidents.The commonly used pipeline leakage detection technology and related equipment have some shortcomings such as complex structure,inaccurate positioning,expensive price and inconvenient online detection,which can not meet the requirements of hydraulic pipeline leakage identification.Therefore,a hydraulic oil leakage detection method based on image recognition was proposed,the specific work accomplished is as follows.(1)This thesis summarizes the common three hydraulic pipeline leakage phenomena and establishes the relevant data set.According to the hydraulic pipeline layout and hydraulic pipeline system scene,the hydraulic pipeline leakage is simulated,and three leakage phenomena are summarized.Based on this,the image data set of hydraulic pipeline leakage is constructed by video image acquisition and image preprocessing.(2)The improved U-net algorithm is proposed to improve segmentation accuracy of pipe joint and pipeline in complex backgrounds.By introducing Mobilenetv3 and LR-ASPP structures,the model volume is reduced to facilitate the later deployment.Afeter that,the attention mechanism and SC adaptive convolution are added to highlight the target feature information and improve the accuracy of the segmented pipe joint and pipeline area.(3)A segmentation-frame difference fusion algorithm for hydraulic pipeline leakage detection is proposed to realize detection and classification of three different leakage phenomena.Firstly,the ROI region is extracted according to the hydraulic pipeline segmentation algorithm.Then,the multi-scale time frame difference method is proposed to fuse the features of piple leakage.Finally,the corresponding judgment rules are summarized to realize the accurate classification of leakage phenomena.Through training and testing,the average false detection rate and missing rate of the frit-frame difference fusion algorithm are 7.59% and 1.59% respectively.(4)The deep learning algorithm YOLOv5 n network is improved to realize hydraulic oil leakage detection under complex pipeline arrangement.By replacing NMS with Soft-NMS,the probability of deleting the overlapping box is reduced,and CBAM attention mechanism and multi-head attention mechanism are introduced to highlight the important features of leakage phenomenon.Through training and testing,the accuracy of the improved YOLOv5 n algorithm is 3.5% higher than that of the original YOLOv5 n algorithm.(5)The GUI interface of hydraulic pipeline leakage detection and recognition is developed,and the hydraulic pipeline leakage detection system is designed.ONNX Runtime is used for model deployment,and Py Qt technology is used to complete the GUI interface design of hydraulic pipeline leakage detection and recognition,and the visualization of hydraulic pipeline leakage detection is realized.