Application Research Of Deep Learning In Surface Defect Detection Of Aluminum Profile

China is a large industrial manufacturing country with many aluminum profile manufacturers,so the quality inspection of aluminum profiles is of great significance.At present,the surface defects of aluminum profiles mainly depend on manual judgment,but this method is inefficient and unstable.With the transformation and upgrading of China's manufacturing industry,companies have higher requirements for product quality,accuracy and inspection speed,and traditional methods are unable to meet inspection requirements.Infrared recognition,ultrasonic flaw detection,eddy current flaw detection,magnetic flux leakage recognition and other single mechanism recognition methods are effective,but the detection speed of these methods is slow and the cost is high.In recent years,surface defect detection methods based on machine vision and deep learning have been developed,but they also have the disadvantages of low robustness,narrow application area,insufficient feature extraction,and limited detection accuracy.In view of the above problems,this subject based on YOLOv3 and Mask R-CNN,and proposes an aluminum profile surface defect detection method based on image fusion and deep learning.Firstly,the original image is preprocessed by the method of grayscale transformation enhancement and spatial domain filtering to obtain the processed image;then draw on the ideas in SLAM to extract and match the original and processed images;afterwards,image fusion is performed to obtain the final processed image;then through K-means algorithm clustering and parameter optimization,and finally use the YOLOv3 model to detect the surface defects of aluminum profiles,through an end-to-end fully convolutional neural network,the input from the original image to the bounding box and the object category and confidence output in the box are completed.At the same time,use Mask R-CNN to conduct experiments,and use the mask to more specifically locate the types and positions of surface defects.In addition,the YOLOv3 model generated by this subject is pruned,transplanted into the Jetson TX2 module,the inspection interface was generated using Qt,finally,the image and video are detected to realize the packaging application of the aluminum profile surface defect detection model.The experimental results show that the method proposed in this subject have an average success rate of 98.33% for surface defect classification detection,which is3.75% higher than the single deep learning method;and the m AP value of validation setis 88.81%,which is 4.18% higher.It has stronger feature extraction and generalization capabilities,which can accurately detect surface defects,classify and locate.