Research On Multi-focus Image Fusion Algorithm Based On Deep Convolutional Network And Its Application In Detection

In the defect detection of industrial vision applications,the quality of the image acquired greatly affects the accuracy of the detection model.Due to the randomness of defects,defects on the surface of the workpiece are always difficult to maintain at a high level.In the process of image shooting,due to the limited depth of field of the camera,objects within the depth of field of the camera can be clearly imaged in the image,while objects outside the depth of field of the camera can only be blurred in the image.However,it is difficult for an existing image acquisition device to directly acquire a complete workpiece defect.The inability to obtain clear imaging of all objects has a great impact on subsequent image detection and recognition.Therefore,the construction of computer-aided detection system based on image fusion can make up for the detection error caused by the limitation of the hardware system and improve the detection accuracy of the system.Based on deep convolutional networks,this paper studies from two aspects of multi-focus image fusion and detection applications.The main research work of this paper is as follows:(1)By improving the encoder-decoder network,a multi-focus image fusion network with pixel correction module is proposed.Firstly,a pixel correction module is introduced into the encoder-decoder network,and the horizontal connection is used to merge the spatial features of the low-level image in the coding network with the semantic features of the high-level image in the decoding network to eliminate the pixel error generated during decoding.Then introduce the structural similarity loss function to reduce the loss of the fusion image on the structural information.Finally,considering the scarcity of existing multi-focus image datasets and the large amount of training data required by deep neural networks,finely labeled image segmentation datasets are used to make training datasets to further improve fusion accuracy.Compared with several excellent fusion methods,the fusion quality of this method is better and the preservation of details is more satisfactory.(2)Use multi-focus image fusion to solve practical problems.In the task of detecting tool gaps in actual industrial workshops,in order to solve the problem that the existing image acquisition system cannot obtain complete information on the surface of the tool under test,a more real-time multi-focus image fusion method is introduced to obtain more complete tool gap microscopic information.For the situation where a large gap cannot be obtained by a single image,an image stitching method based on envelopes is introduced to better display the large gap through a stitched image.Finally,an original edge line of the tool is fitted according to the image envelope,and combined with the actual envelope to complete the precise detection of the gap.The experimental results show that this method has a good recognition effect on the tool gaps that cannot be photographed clearly.