Image Recognition System Of Engineering Vehicle Equipment Based On Deep Learning

At present,deep learning technology has been widely used in computer vision related fields,and it has spawned actual market products in various application scenarios such as medical imaging,identity recognition,intelligent transportation,and image synthesis.However,with the gradual deepening of deep learning technology theoretical research,how to use advanced theoretical technology to design and manufacture practical market products,so as to bring practical convenience to people’s social production and life,still requires our continuous exploration and research.This thesis designs a data management system that combines deep learning technology and Internet of Things architecture,which can be used to automatically classify a large number of engineering vehicle equipment images collected manually during engineering vehicle maintenance,and focuses on the image classification model used in the system And the structure and characteristics of each functional module of the system.In the training process of image classification model,better training effect usually depends on massive data resources and fine network structure.On the one hand,in the process of acquiring data resources,if you only rely on human acquisition,it will not only consume a lot of time and effort,but also it is difficult to collect large-scale data resources for certain scenarios.To solve this problem,the usual approach is to Before training,the data set is enhanced,but it is necessary to design a reasonable data enhancement method for specific problems.Based on the confrontation generation network,this thesis proposes an image generation algorithm suitable for the image set of engineering vehicle equipment.The network parameters are optimized and used The two learning networks of generator and encoder fight against each other to produce clear and diverse fake images.On the other hand,in the construction of image recognition network structure,complex and detailed network structure training often means a huge amount of calculation and lengthy training time,and often accompanied by serious overfitting problems,so this article uses migration-based The training method of learning and model fine-tuning,by inheriting the advantages of the weight matrix of the existing mature network,freezing most of the network fine-tuning some network parameters,avoiding excessively high computing costs,and generating image classification with good recognition effects model.In addition to the theoretical research of the algorithm,this article also focuses on how to use the existing theoretical research results to build a concrete system with practical use value.This thesis uses the trained image recognition model,combined with the Internet of Vehicles technology of multi-functional engineering vehicles,and actually builds an automatic classification system for engineering on-board equipment.This system includes the collection and reporting from the terminal to the identification and storage of the server.Two functional modules,using a reporting strategy based on storage and polling mechanisms to ensure the reliability of large-scale data reporting,and an image classification strategy based on content weights to increase the accuracy of image identification.The system can be used to automatically store and classify manually collected engineering vehicle equipment images during engineering equipment maintenance,which can greatly enhance the data value of user data assets.