Research On Key Technology Of Thermal Power Plant Recognition Based On Remote Sensing Images

The power generation process of thermal power plants is often accompanied by the burning of a large amount of coal,producing various harmful compounds and smoke and other substances,which seriously damages the public ecological environment.Therefore,it can provide technical support for environmental monitoring that the advantage of the wide range of visible light remote sensing image detection is used to identify the thermal power plants and the advantage of high efficiency,long distance of infrared remote sensing images is used to segment the deformable targets such as smoke and harmful compounds released by thermal power plants.This paper focuses on the key technologies of thermal power plant target recognition based on remote sensing images,and realizes the recognition of thermal power plant targets and segmentation of deformable targets.The main research contents and achievements are as follows.Aiming at the current deep learning’s low accuracy in detecting and identifying fixed buildings based on remote sensing images,a multi-label thermal power plant recognition method based on graph convolution network is studied and implemented.First,a convolutional neural network is used to learn image features as a basic model;second,the correlation matrix of the multi-label graph convolutional network is used to mine the co-occurrence pattern between image tags,and the dependency relationship of the tags is modeled;then,graph convolutional networks is used to construct directed graphs on object labels and learn interdependent object classifiers;finally,object classifiers is applied to image descriptions extracted by convolutional neural networks,traditional multi-label classification losses is used to train the entire network to get the multi-label prediction of the image.Through multiple sets of experiments,it is found that the method in this paper can achieve the recognition of thermal power plant targets through multi-label combination.Aiming at the problem that the existing infrared image gas detection method is difficult to accurately determine the gas range for the deformable target,a deformable target segmentation method based on a mask score instance segmentation network is studied and implemented.First,a new mask scoring mechanism is designed,which uses the pixel overlap rate between the predicted mask and the matched real mask as the mask score;second,the level-overlap prediction branch is added to the original Mask R-CNN instance segmentation network,which recalculates the mask score;finally,the network is trained end-to-end on the constructed gas dataset to improve the quality and integrity of the gas target instance mask.By testing multiple sets of infrared long wave gas data,it is found that the segmentation effect of the proposed method is superior to other classic algorithms.Aiming at the problems of inadequate utilization of computing resources and Inefficient information processing due to the massive data required by the existing deep learning training,a cloud computing-based information processing and analysis framework was designed and implemented to complete thermal power plant target recognition task and deformable target segmentation task.The framework uses Docker containers as the unit computing platform,and uses Kubernetes container orchestration technology as the platform for managing containers.A private cloud computing platform based on research on key technology of target recognition in thermal power plant of the container cluster system is built,which allows administrators to manage cluster resources in the form of containers,and allows the user to submit the configured environment and algorithm of multi-label recognition in thermal power plant and deformable target segmentation as a mirror in the container,making full use of GPU cluster resources to achieve the recognition of thermal power plant targets and the segmentation of deformable targets..