Research On Hyperspectral Image Of High Conse Quence Area Oil And Gas Pipeline Classification Method Based On Dual Improved Generative Adv Ersarial Networks

With the development of information science and technology,the development of remote sensing technology is also deepening,and the multi-dimensional information of hyperspectral remote sensing image also presents more abundant geographical environment details.In the actual oilfield industrial site,oil and gas transmission pipeline bears an important safety responsibility.When the oil and gas transmission pipeline accident occur,it will threaten the safety of people's lives and property,pollute the natural ecological environment,and cause great damage to such areas.The areas prone to such safety risks are called high consequence areas.Therefore,strict control of high consequence areas is the top priority Heavy.Nowadays,UAV technology and satellite remote sensing technology are becoming more and more mature.Through such sophisticated instruments,we can accurately control the remote sensing image and ground information around the oil and gas transmission pipeline in the oil field industrial site,accurately locate the surrounding natural environment,residential areas and other geographic information,once there are security risks,we can grasp the problems in real time,and then take timely remedial measures.However,the oil and gas pipeline is usually continuous,and the real-time monitoring of high consequence areas at all levels cannot avoid consuming too much manpower and financial resources.Therefore,the reasonable application of remote sensing images of high consequence areas plays an important role in the control of ground information and future geographic planning of high consequence areas.It is of practical significance to classify the remote sensing images of high consequence areas of oil and gas pipelines.Along with the continuous progress of deep learning technology,the paper provides a new research idea to deal with image classification problem by using the generation countermeasure network technology.The generation countermeasure network not only has a complete network to correct,but also has better classification effect than the traditional algorithm.Therefore,the paper mainly studies the classification method of remote sensing image based on the generation countermeasure network Research.The main work of the paper is as follows:Firstly,in view of the problem of the over distance between classes and the class spacing among the same categories in hyperspectral remote sensing image information,the middle and high-level feature layer method is introduced.In the paper,the middle and high-level feature layer is introduced into discriminator D of GAN network,which can not only effectively alleviate the problem of over fitting,but also pixel the high information depth in the image,and can effectively alleviate the problem of over fitting,and pixel the high information depth in the image The advanced features of hyperspectral remote sensing images are extracted further,and the final image classification accuracy is improved and the classification results are improved significantly.Secondly,in view of the problem of low data and redundancy of hyperspectral remote sensing images,the paper uses the generation ability of generation counter network GAN to expand the data set processing of the generated samples.With the cooperation of the built-in convolutional neural network,the random generation ability of GAN can be effectively converged,and the samples can be generated as close to the images in the real data of the target as possible,which enriches the diversity of data sets.For the purpose of the paper,the data sets are enriched the following network training and iterative tuning play an active role.Thirdly,in order to further improve the classification accuracy of multidimensional data,the label mechanism is introduced into generator,and the corresponding class tags are generated to assist classification while generating adversary samples.It not only relieves the pressure of discriminator,makes the network robust,but also alleviates the disorder distribution in the training iteration process,further improves the classification accuracy and makes the students generate the label The integrator is more suitable for the category of label to complete the generation task,which provides a good help for the subsequent feature extraction and classification.Finally,the performance evaluation is carried out by using the Overall Accuracy OA,Average Accuracy AA and Kappa coefficient K classification index.By comparing with the classical image classification method,the superiority and universality of the improved algorithm are verified.The method is applied to the remote sensing image of high consequence area in the actual oilfield industry field,and the classification effect of the improved algorithm is analyzed.