Research On Multi-Type And Multi-Target Recognition And Binocular Positioning For Domestic Ceramics

Domestic ceramics are all kinds of ceramic products for daily life,It mainly includes tableware,tea sets and other ceramic utensils,etc.It has the characteristics of many types and big differences in shape.In order to achieve machine sorting requirements,A multitype multi-target recognition and binocular positioning method based on deep learning domestic ceramics is proposed,The main research content and work are as follows:1.The YOLOv3 netword model had been researched and improved.Under the framework of Tensor Flow,a multi-target detection network had built basd on using the improved model of the deep learning.The recognition of the selected six types of domestic ceramics had been realized.Experiments show that the network has strong learning ability and high type recognition rate.2.The lightweight had been researched for target recognition algorithm,The research had used the K-means algorithm to re-cluster the anchor box to generate a prediction box size that meets the target detection of domestic ceramic types,Proposed and implemented a lightweight model Mobile Net v3 to replace the Dark Net-53 feature extraction network in the original YOLOv3,got a lightweight YOLOv3-M3 network model,Compared with other network models,The recognition accuracy of this model is significantly better than other models in all selected household ceramic types,The MAP value reached 90.27%,For the input 416*416 video stream,The detection speed is above 18 fps,and the model occupies less memory.3.The binocular stereo vision platform had been built and calibrated the system.Aiming at the problem of weak texture matching of domestic ceramics,the research had used a SGBM based on semi-local matching algorithm,the image had Graying processing,through the lightweight model had been identify the domestic ceramic area in the image,got the position of the target center point,According to the principle of binocular vision positioning,realized the positioning of the target center point of domestic ceramics,the average positioning error is 7.405 mm and the error rate is 0.92%,which can meet the sorting requirements of domestic ceramics.