Disc disk blade surface defect detection method based on improved Faster RCNN algorithm

• Main Authors: SUN RUIRUI, CHENG DAQUAN, WANG XINGXING, ZHANG RANGYONG, CHENG GUANGHE
• Format: Patent
• Language: Chinese; English
• Published: 31.01.2023
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; IMAGE DATA PROCESSING OR GENERATION, IN GENERAL; PHYSICS

The invention belongs to the technical field of image processing detection and computer vision, and more specifically relates to a disc harrow blade surface defect detection method based on an improved Faster RCNN algorithm, and the method specifically comprises the steps: S1, collecting disc harrow blade images on a production line through a fixed device, and picking out images with defects; s2, labeling, dividing and enhancing the collected pictures, and determining a training set and a verification set; s3, the model based on the Faster RCNN algorithm is improved, wherein a K-means clustering method is adopted to automatically generate the proportion of anchors; resnet50 + FPN (Fabry Perot Network) is selected; the ROI (Region of Interest) Align is selected to replace the ROI An EIOU NMS designed based on a generalized intersection-to-parallel ratio EIOU is used; s4, improving a loss function during training, and adding a CA attention mechanism into the model; s5, carrying out the training of the Faster RC