A Study On The Detection Technology Of Illegal Buildings In Fixed-point Monitoring Mode Considering Meteorological Disturbance

The supervision of illegal buildings is one of the priorities in the development of new urbanization in China.In view of the current problems of low automation degree and the susceptibility to meteorological factors in illegal building supervision methods,this paper is devoted to the study of key technologies for automatic identification of illegal buildings under complex weather conditions.The main research contents are as follows:(1)An weather recognition algorithm for image combining SVM Chains and MLKNN is proposed.First,by using the differences in the edge,brightness,and distribution of the sky area and the non-sky area,combined with the Canny edge detection algorithm,an effective segmentation of the outdoor image sky area was achieved;then,by analyzing the image quality degradation and color changes under different weather conditions Characteristics such as brightness,color,contrast,saturation,dark channel,and rain and snow noise characteristics were designed and extracted;finally,by combining the SVM Chains and ML-KNN algorithm,the recognition of outdoor image mixed weather conditions was realized,so as to provide reliable weather information for subsequent anti-weather interference algorithms.Experiments show that the accuracy of this algorithm to recognize multiple weather conditions in outdoor surveillance images is better than 90%.(2)A dark channel haze removal algorithm based on corrected atmospheric light value and optimized transmittance is overcome.First,a haze density description algorithm based on the statistical characteristics of natural scenes is proposed,which realizes the estimation of the haze density in the monitoring image;then,Aiming at the problems of low brightness and color distortion of the defogging result of the dark channel prior defogging algorithm,the region recursive method is used to modify the atmospheric light value,and the fog-free image in the same region is used to optimize the transmittance,thereby improving the defogging ability of the original dark channel algorithm.Experiments show that the improved algorithm is superior to the original algorithm in direct vision and quantization indicators such as SSIM and peak signal-tonoise ratio.(3)A CNN-based illegal building recognition algorithm based on the comparison of old and new relative images is studied.Firstly,to solve the problem of small target recognition in large scenes,an algorithm for locating and extracting suspiciously changing regions of old and new phase images based on color space distance is proposed;then,a similarity discrimination algorithm for image structure based on transfer learning is proposed.The algorithm effectively eliminates the pseudo-change areas;finally,based on the Inception V3 network model,the buildings in the changing area are identified,and the final illegal building detection results are obtained.Experiments show that this algorithm accurately and effectively realizes the identification and location of illegal buildings in fixed-point monitoring mode,with an accuracy rate of 89.7% and a recall rate of 96.2%.