Research And Application Of Automobile Door Plate Solder Joint Recognition Algorithm Based On Deep Learning

At present,the traditional welding robots has been widely used in the welding production line of interior and exterior decorative parts of automobile door panels in China.The application of traditional welding robots in welding production lines can improve production efficiency,but traditional welding robots generally use manual teaching methods in the process of identifying welding.When a product is changed,manual teaching planning is required,which reduces the efficiency of the automotive welding line.In order to improve production efficiency,this paper studies the target detection algorithm based on deep learning for real-time recognition of automotive door panel solder joints.The main research content of this paper is the target detection algorithm based on deep learning.Before delving into the target detection algorithm based on deep learning,we first introduce the traditional target detection method.The process of the traditional target detection method is generally: preprocessing the image,then extracting the feature of the preprocessed image,and finally classifying the extracted features by using the classifier.Then the related theoretical knowledge of neural network and deep learning is introduced,which paves the way for the research of target detection algorithm based on deep learning.It mainly includes the structure of convolutional neural networks,common convolutional neural network architecture,regularization methods in deep learning and optimization algorithms.The commonly used convolutional neural network architecture can improve the performance of target detection.The regularization method in deep learning can solve the over-fitting problem in deep learning.The optimization algorithm in deep learning can be used to improve the efficiency of neural network training.Finally,the target detection algorithm based on deep learning is studied and compared experiments.The target detection algorithms based on deep learning are divided into two categories.The first type is the target detection method based on the candidate region.The target detection method based on the candidate region originated from the R-CNN algorithm.Because the R-CNN algorithm has the disadvantage of slow detection speed,the Fast R-CNN algorithm is improved,and the Fast R-CNN algorithm is further improved with the Faster R-CNN algorithm.The second type is the target detection algorithm based on regression method,namely YOLO algorithm.Because the YOLO algorithm has poor recognition effect on small targets,the improved YOLO algorithm is proposed.The experimental results of the Faster R-CNN algorithm and the improved YOLO algorithm show that the two algorithms have higher recognition rate for the automotive door panel solder joints,and the accuracy of the Faster R-CNN algorithm is over 80%,and the improved YOLO algorithm is not only higher in accuracy than the Faster R-CNN algorithm.And the detection time is much faster than the Faster R-CNN algorithm,which can meet the real-time requirements of industrial production.