Research Of Fault Indicator State Image Recognition Approach Based On Convolutional Neural Networks

With the development of machine vision technology,more and more digital image processing techniques are used to analyze and identify the quality of products,In order to realize the intelligent detection of fault indicator products which produced in the factory.Fault indicator state recognition based on convolution neural networks(CNN)is studied,which can effectively solve the problem of product testing.In this paper,based on fault indicators intelligence detection task,the construction of recognition system,the improvement of CNN model and experimental verification are studied.First,the situation of the product detection is analyzed,the image acquisition system is constructed and the original video image is collected.Then,this paper designs the recognition algorithm flow,and experimental results show the feasibility that CNN is used directly for the images.But analyzing the results,which find out the problems in CNN.Next,image quality problems are existed,include low resolution,uneven lighting,color cast.The pretreatment is enhanced and highlight removal,which reducing the influence of various factors on the recognition.Further the threshold,edge detection and clustering method for image segmentation based on the experimental data.By image translation or rotation to extend data set,improving the recognition performance of small sample of CNN.For the traditional CNN model’s robust problem,from the network structure,proposes a multi-scale CNN model,through the experiments verify the robustness of the method.For the convergence time is too long and the recognition rate is too low of the traditional CNN,great correlation between the each kernel function resulted in redundancy,proposes the method of initializing the first-layer kernel function by wavelet transform,experimental results show it can shorten convergence time,The combination of these two improved methods,compared with the traditional CNN,the recognition rate increased by 7.28%,reaching 96.32%.Finally,Faster R-CNN model set detection and recognition in one has also been applied to the fault indicator status recognition task of this paper,the experimental results show that it can effectively solve the problem.