Research On Automatic Detection Of Illegal Buildings Based On Digital Image Processing

In recent years,with the booming real estate industry and the national housing renovation plan,the number of illegal buildings has increased,which has brought great challenges to staff.The rise of small drone technology has improved the efficiency of illegal building detection,but this way is rely on professionals to operate drones to identify illegal buildings.Establishing a set of automatic detection methods for illegal buildings can improve work efficiency and detection accuracy.The traditional digital image processing technology has high precision in detecting pixel value differences,but it cannot perform semantic segmentation and identification of pixels.Fully convolutional networks can achieve semantic segmentation and identification of pixels.This thesis proposes an automatic threshold filtering algorithm that combines digital image processing technology with full convolutional network technology to establish a set of illegal building automatic detection methods.The data source collection of illegal buildings is obtained through the drone fixed time and route shooting.Use the SURF algorithm to calculate the common feature points of the comparison pictures and to filer the feature points that meet the requirements.Solve the coincidence matrix for image registration and use the proposed dynamic threshold algorithm to detect image change areas and then pass change areas picture into the fully convolutional network model to identify.Finally,the illegal building area is automatically marked based on the recognition result.Use the Django framework to encode the change detection method of this article to implement the interface form,which is convenient for fully convolutional network to call.In order to test the feasibility of the method proposed in this thesis,a quantitative accuracy evaluation method was used in the experimental to measure experimental results.Evaluation methods include correct rate,false rate,miss rate,and average error rate.Research indicates that the automatic detection method of illegal buildings in this thesis is improved by12.4% compared with the detection method of support vector machine two classification method and compared with morphological building index method that the accuracy rate is increased by 28.1%.This method has important application value.