Research On Recognition And Positioning Of Industrial Parts Based On Artificial Neural Network

Due to the advancement of society and the application of science and technology,deep learning technology has become the main means of processing industrial parts identification,and has gradually begun to be applied to many areas of life.However,in the actual processing of industrial parts classification,there will be many types of parts that are too similar in appearance,each product is shielded from each other,the features of the parts themselves are less or less prominent,and other complex condition factors interfere.This reduces the accuracy of YOLO and SSD target recognition algorithms.Therefore,this article addresses the problems of low detection efficiency,poor detection effect,and high miss-detection rate of industrial parts under the abovementioned specific conditions.Based on deep learning,it is carried out from three aspects: the collection of original data,the preprocessing and construction of the data set,and the improvement of the model.Research and optimize the YOLOv3 algorithm,improve the detection and recognition performance of the occluded part and the small target part,and achieve a higher accuracy rate on the self-collected data sets of more than a dozen types of parts.In order to improve the quality of data sets,this paper collects a large number of target images under the above specific conditions by adjusting the target angle,illumination and background,aiming at the specific scene where the detection targets are industrial parts with similar shapes,mutual occlusion and unobvious features.At the same time,considering the data requirements suitable for deep learning,the target part data set is preprocessed to complete the construction of the data set needed for this project.Label Img software is used to mark the required data set,and K-means++method is used to re-cluster the areas with dense industrial parts,and the size of anchor frame suitable for this subject is obtained,which improves the detection performance of the model for the area frame where the target is located.To solve the problem of missing detection or low detection accuracy caused by insufficient feature extraction ability of original YOLOv3,in order to improve the recognition and detection ability of the model for target parts,a YOLOv3-SPP-4l algorithm is proposed for complex parts detection.Firstly,Focal Loss algorithm is used to optimize the loss function,and the corresponding parameters are changed in the training process to improve the recognition accuracy of effective samples of small parts.Secondly,by improving the original network structure of YOLOv3,using SPP module and increasing the feature extraction scale,the fusion between the four feature scales is realized,and the ability of the model to extract target features is enhanced;At last,based on pytorch framework,the improved YOLOv3 algorithm proposed in this paper is verified by experiments.The research results show that compared with the original YOLOv3,the improved model not only improves the detection accuracy of small targets,but also reduces the error rate when the targets are stacked and the features are not obvious.Therefore,the improved method in this paper can be used for reference in the identification of parts under special unstructured conditions.