"Machine vision" and "positioning accuracy" are two aspects that affect flexible automatic production line in intelligent manufacturing. Machine vision gives flexible production perception ability. In practice, it can be found that machine vision is an important part of realizing "perception" in the automatic production line of products. Machine vision is to use machines instead of human eyes to do identification, measurement, detection and semantic understanding.
With the development of artificial intelligence technology, the application of machine vision in factories is increasing. Many problems that traditional machine vision can't deal with are solved. Image recognition mainly includes feature extraction and classification recognition. The traditional extracted features are the visual features of the image bottom layer, and need to have a certain professional knowledge personnel to design and select the features. This kind of manually designed features need to be verified after a lot of verification to prove its effectiveness for a certain recognition task, which also limits the application of image recognition technology to a certain extent.
In view of the limitations of traditional image processing algorithms, for example, when making a two-dimensional code, we first need to know the characteristics of the two-dimensional code, then do image preprocessing, find the problem of the two-dimensional code, and then solve the data of the two-dimensional code. There are many defects in this way.
In the process of location based on the features of QR code, it is difficult to identify the QR code with disordered background, uneven illumination, perspective deformation and poor printing quality. Because the background is disordered and irregular, it can not effectively distinguish the features of the target. In addition, the feature of the target changes significantly in a specific scene.
Therefore, it is necessary to improve the ability of feature extraction and generalization. When deep learning is applied to machine vision, the deep learning convolutional neural network is used to recognize two-dimensional code. The deep neural network simulates human visual process. The front layer only perceives the edge contour, and different neurons in the back-end layer "excite" to generate local features, and then to generate panoramic images. The mechanism of automatic extraction of image features and the processing process similar to human brain greatly improve the effect. After that, the two-dimensional code can be correctly recognized, which greatly improves the success rate and efficiency of machine vision recognition.