Research On Environmental Testing Of Chemical Plants Based On Aerial Photography Of Unmanned Aerial Vehicles

With the implementation of the national "14th Five-Year Plan" energy conservation and emission reduction policy,there are higher requirements for environmental testing of chemical plants.It is particularly important for the detection of leakage in chemical plants.The chemical plant environment is disturbed by a variety of factors,with a variety of targets and difficult detection.At present,the traditional method of detection is mainly time-consuming and labor-intensive,and it is particularly important to find an efficient and simple detection method.Based on the advantages of unmanned aerial vehicles in taking images,this paper completes the environmental detection of chemical plants based on drones and drone aerial photography,that is,the target detection of smoke and sewage outlets through images,the color classification of sewage images and smoke images,and finally the detection of air quality and temperature and humidity through intelligent sensing equipment.The main work is as follows:(1)For the target detection of aerial images taken by unmanned aerial vehicles,YOLOv4 has been improved to obtain the ME-YOLOv4 target detection network to achieve target detection of sewage outlets and chimney smoke in the chemical plant environment.The network uses deeply separable convolution and Mobile Netv3 to improve the network structure and increase detection speed;methods such as image preprocessing,data augmentation,ECANet attention mechanism,and clustering are used to improve detection accuracy.On the VOC dataset,the detection accuracy reached 82.5%.On the smoke and sewage outlet dataset,the detection accuracy rate is 87.65%.The detection speed is 49 FPS,which is 2.45 times higher than the YOLOv4 detection speed and 1.6 times higher than the YOLOv5 detection speed.(2)Aiming at the problem of color classification of sewage and smoke images,the results of the target detection network are first used to intercept the detected targets of smoke and sewage,and the color classification data set is obtained.Through data augmentation and normalization processing,the impact of images on the detection results is reduced,and then the classification network Res Net50 is used to classify smoke and sewage by color characteristics,with smoke color classification TOP-1 being 96.27 and sewage color classification TOP-1 being 99.92.It is superior to VGG16 and Mobile Netv2 in classification accuracy.(3)For air quality detection,the UAV airborne embedded sensing device is used to detect air gas quality,PM2.5 and temperature and humidity,complete the construction of the UAV air quality detection system,and test the data detected by the sensor,and finally upload the data through the 4G module to Alibaba Cloud to achieve data preservation and build the data display and data analysis interface through Io T Studio.