Research And Implementation Of Small-Scale Waters Float Recognition Method Based On Feature Fusion

In rivers,lakes,reservoirs and other waters,the floating object such as various kinds of production and living garbage damages the water ecological environment seriously.In order to ensure the cleanliness of the water,it is very important to clean up the surface floats.At present,the cleaning of surface floats mainly adopts the method of manual operation,which has the problem of high cost in both human and material.In order to remedy efficiency of the traditional manual cleaning mode,it is necessary to intelligently clean the floats,and it is particularly important to identify the floats independently in the process of intelligent cleaning.For this reason,this thesis proposes a method to identify surface floats,which has important engineering application value in the field of environmental protection.The surface floating object identification method is mainly divided into three steps: water surface image preprocessing,surface floats feature extraction and pattern recognition classification.The key and difficult points of this thesis are image preprocessing and establishing an effective target feature model.In order to solve the problem that the foreground target,caused by the complex water surface environment in the image preprocessing stage,is difficult to segment,a background-based priori image segmentation method is proposed.Firstly,the image is denoised by median filtering and sharpening filtering.Secondly,the water surface background model is established by using the improved mixture Gaussian background modeling method.Then,the saliency region is extracted according to the prior background model to obtain significant gray image.Finally,the improved maximum inter-class variance method is used to segment the foreground target region to obtain a binary image,and the binary image is secondarily segmented by the significant grayscale image.Aiming at the problem of target feature modeling,a method combining color feature model,edge and region feature model is proposed.The color feature model is constructed by using the color moment and the color difference between the target and the background area,and the regional feature model of the target is constructed by the compactness,appearance ratio,concavity and invariant moment of the target image,while the edge contour feature model is constructed by using the chain code difference feature and the second-order difference feature of the boundary to the target edge distance.For the feature recognition problem,a support vector machine(SVM)is proposed to identify the target feature and determine whether the target is a float or not.The experimental results show that the background-based priori image segmentation method effectively highlights the foreground target region and segments the foreground target.The proposed feature model can achieve a success rate of 82%.