The Research On Recognition Technology Of Gas Station Equipment On Convolutional Neural Network

At present,the per capita ownership of automobiles is increasing year by year,and the requirements for the planning,construction,operation and maintenance of gas station equipment are becoming higher and higher.With the rapid development of artificial intelligence,target detection and recognition methods based on deep learning are gradually applied to various industry fields.In China,how to use relevant theories and technologies to carry out automatic safety inspections of internal facilities of gas stations to improve work efficiency and save human resources and costs has also attracted people’s attention.This article focuses on the gas station equipment identification method based on the deep convolutional neural network based on the relevant specifications of gas station equipment safety,and has completed the following aspects of work:1.By referring to literatures in target research and identification related research fields,as well as relevant standards for the layout of internal facilities during the construction and operation of gas stations,according to the specific actual needs of equipment safety inspections,the traditional identification methods and deep learning nerves are analyzed The characteristics of the network structure compare the performance and applicability of related methods.Aiming at the characteristics of the scene in the complex environment of the gas station,in order to solve the problem of multi-object recognition,the Mask RCNN algorithm is mainly discussed,and the advantages and disadvantages of the Mask RCNN neural network structure are analyzed,and the main research content and methods used in this article are clarified.2.Aiming at the problem of uneven segmentation effect in Mask RCNN network,a multi-convolution branch processing structure is proposed.The structure is based on the Mask RCNN network framework,and the Mask Io U module is added to increase the segmentation accuracy of the mask by increasing the convolution layer and the fully connected layer,thereby improving the overall recognition effect.On the COCO2017 public data set,an experimental verification work was carried out with an improved network model.The experimental results show that the multi-convolution branch processing structure based on Mask RCNN network greatly improves the segmentation accuracy,and the segmentation effect and recognition accuracy are more excellent.On this basis,the recognition accuracy has risen to a new level.3.According to the regulations in "Code for Design and Construction of Automobile Gas Station GB50156-2012",the design of the gas station scene experiment sample set and the collection of image data were carried out.For different equipment in the gas station,onsite photo shooting,sorting and enhancement were carried out in different environments.Then proceed with data preprocessing,and manually label the collected data with VIA plugin,thus completing the construction of the data set.4.Based on the data set produced in this paper,the improved multi-convolution branch processing structure model was used for training and testing.The feature pyramid network method was used to fuse context features,and feature images at different scales could be learned.Based on this,in the region It is recommended to generate candidate windows in the network,and complete the classification and segmentation process at the same time,to achieve the integrated operation of detection,recognition,classification and segmentation of the live image of the gas station.Through comparative experiments,it is verified that the recognition performance,learning ability,stability and speed of the method in this paper all have good performance.The results of the test using real-world data also have better performance in accuracy and efficiency.