Research On Detection Method Of Sewage Outfalls In Uav Aerial Image Based On Deep Learning

Survey of land-based discharge outlets refers to the periodic inventory and statistics of the discharge outlets into the sea and rivers,which is an important means for the ecological and environmental departments to grasp the actual situation of land-based discharge and curb illegal discharge.In order to improve the efficiency of census,UAV aerial photography has been introduced into land-based discharge outlets for census in recent years.However,the interpretation of aerial images mainly relies on manual methods.In this thesis,based on the deep learning model,an intelligent detection method for land-based sewage outlets is proposed in view of the characteristics of small targets,small samples,complex background and low recognition of sewage outlets in aerial photography of UAV.The main work of this thesis is as follows.First,we propose a small-target-detection method for land source outfalls.By improving the feature extraction network of fast RCNN network,the layer with better effect on small target extraction is selected,and the feature fusion is carried out after screening.The fused feature can retain the small target information as much as possible.The improved network model can achieve 0.7010 detection accuracy(AP)and 95%recall rate for sewage dam.Second,we present a small-sample-detection method for land source outfalls.Aiming at the problem of insufficient data of sewage outlets such as sewage pipes,two methods of image transformation enhancement and deep convolution generation are used to expand the number of samples,which reduce the over fitting phenomenon in small sample training;and the problem of insufficient positive sample information in small sample data set is solved by anchor frame clustering,which greatly improves the efficiency of the method The results show that the model is effective for small sample target detection.Third,we design a system SOIS of aerial image sewage outfall intelligent identification.In order to apply the sewage outfall detection method to engineering practice,this thesis builds the aerial sewage outfall intelligent identification system SOIS based on the flash framework,ncnn framework and Android development technology.The system integrates model deployment,model training and target detection,and can accurately output the longitude and latitude coordinates of the sewage outfall according to the flight parameters of UAV.The design and implementation of the system reflects the practicability of the sewage outfall detection method.The method of UAV aerial sewage outfall detection based on deep learning proposed in this thesis can greatly improve the detection efficiency of land source sewage outfall,and improve the utilization rate of aerial image;the design method of aerial sewage outfall intelligent recognition system also provides a research basis for the application of target detection method in engineering practice.