Industrial product label number identification method based on convolutional neural network

Abstract

The invention relates to the technical field of industrial detection. The invention particularly relates to an industrial product label number identification method based on a convolutional neural network. The method comprises the steps of training and establishing a model and recognizing a label number; completing an image recognition task by utilizing a deep neural network VGG-16; and effectively improving the accuracy of distinguishing a faulty mold by accurately classifying workpiece digital labels. Compared with a traditional image classification method, the method has the advantages thatdeep learning is adopted; the accuracy is high; and the robustness is high; compression and reasoning acceleration are performed on the VGG-16 by using a channel pruning method in neural network pruning. According to the method, the trained model can be compressed by about 25 times, and the floating point operation operation is reduced by about 4 times, so that the model can be deployed into theembedded equipment, the cost budget is greatly reduced compared with a cloud computing mode, meanwhile, the delay is avoided, and the analysis data can be processed in real time.

Description

technical field [0001] The invention relates to the technical field of industrial detection, in particular to a method for identifying industrial product label numbers based on a convolutional neural network. Background technique [0002] The mobile phone camera bracket is a tiny industrial part that fixes the mobile phone camera we use every day. The part is produced by a special mold. During the production process, various failures will occur in the machine that produces the workpiece, resulting in the produced workpieces not meeting the standards, and a large number of workpieces being scrapped, which directly affects the production efficiency of the factory and wastes a lot of production costs. Therefore, in order to ensure the normal operation of the factory In production, technicians need to perform fault detection and diagnosis on the produced workpieces. In the workpiece fault detection, the manual detection method is generally used for fault detection. It is diffic...

AI Technical Summary

Problems solved by technology

For image classification tasks, traditional image processing methods can be used to extract image features, and then use support vector machines as classifiers for classification. However, the disadvantages of traditional methods are low accuracy, poor stability, and slow image recognition speed.

[0004] In addition, usually we have to increase the level of the network or the number of filters in order to achieve higher accuracy of the model, but doing so will make the trained model consume a lot of computing resources, storage resources and power

For example, VGG16, an efficient feature extraction convolutional neural network, contains 130 million parameters, occupies more than 500MB of storage space, and generates 15.6 billion FLOPs when inferring a picture with a resolution of 224×224 ( Floating-point operation operands), the cost is high, making it difficult for ordinary small enterprises to apply it to industrial production

Images