Research On Image Detection Method Based On Multi-feature Fusion And Classification

In recent years,landslides,mudslides,and other geological disasters have greatly affected people’s daily activities and lives.With the development of space remote sensing and unmanned aerial vehicle technology,ground changes and transportation facilities can be monitored in real-time around the clock,providing powerful technical support for people to monitor and prevent disasters in real-time,maximizing the protection of people’s lives and property.The image detection method can quickly extract the changes of monitoring objects at different times,which has been widely used in geological disaster monitoring,key facilities’ damage monitoring,and many other fields.According to the characteristics of multi-temporal remote sensing image and bridge crack image,this paper extracted multiple image features and studied the image detection method based on multi-feature fusion and classification.This paper proposes a remote sensing image change detection method based on multi-feature fusion and a bridge crack image detection algorithm combining self-supervised learning and ensemble learning.These two algorithms make full use of texture features,neighborhood features,edge feature and color features of the image to suppress image noise information,improve the detection accuracy of the algorithm,and can extract the image target information accurately and efficiently.The main research work of this paper is as follows:(1)A multi-feature fusion remote sensing image change detection algorithm is proposed based on Principal Component Analysis(PCA)information entropy.The algorithm makes full use of texture features and neighborhood features of remote sensing images and can suppress speckle noise well.Firstly,texture features of images are calculated by Gray-level Co-occurrence Matrix(GLCM),five texture feature difference images with spatial texture features are constructed respectively,and three neighborhood feature difference images with neighborhood features are constructed by introducing neighborhood log-ratio operator.Then fusion of multiple feature difference images based on PCA information entropy by entropy weighting method to obtain a master feature difference map that retains more feature information.Finally,the change detection results were obtained by segmenting the main feature difference image using the Fuzzy C-means(FCM)clustering.The SAR(Synthetic Aperture Radar)image datasets and high-resolution remote sensing image datasets are compared for different change detection algorithms.Experimental results show that compared with other mainstream change detection methods,the method significantly improves the accuracy of change detection results and has better performance.(2)A bridge crack detection algorithm is proposed based on self-supervised learning and ensemble learning.The algorithm consists of two stages: unsupervised initial coarse segmentation of cracks and supervised refinement of segmentation of cracks.The result of the unsupervised initial segmentation is used as the label for supervised segmentation,thus avoiding the labor cost of producing label data.Firstly,the crack image is segmented by local gray consistency,and a binary image of the crack region with much noise is obtained.Then,tensor voting was carried out for the target region,and the structural information of the fracture was deduced from the fractured image under the noise,and the curve and point significance information of the fracture were obtained.Finally,morphological refinement is used to obtain the initial coarse segmentation fractures from the feature significance results.The method extracts texture features,edge feature,and color features of crack images and then inputs multiple features and coarsely segmented cracks into a Random Forest Gaussian Naive Bayes(RF＿GNB)ensemble classifier based on the stacking ensemble method for training and prediction.The abnormal data processing method based on the One-Class Support Vector Machine(OCSVM)was used to deal with the problem of unbalanced fracture sample data,and Random Forest(RF)was used to screen multiple features to reduce feature dimensions.Finally,compared with other crack detection algorithms,the experimental results show that this algorithm can better extract bridge cracks,and have a better detection effect and stronger generalization ability.The proposed method makes full use of the multi-feature information of the image,and not only can obtain better detection accuracy,but also can effectively reduce the detection cost,and has a broad application prospect.The next step will be to consider the influence of shape features on image detection,and study the changing flow of feature classes in remotely sensed images,damage level assessment of fracture images,and types of disease.