Semantic Segmentation Of Street Scene Based On Random Forest Algorithm

Random forest algorithm has many advantages in image processing,such as fast processing speed,high accuracy of pixel classification and easy implementation.It can be used in image classification,image segmentation,detection and recognition and other fields.At present,there are many semantics segmentation methods.The traditional semantics segmentation methods have low segmentation accuracy,while the existing deep learning methods are too complex.Based on this,two semantic segmentation methods combining random forest with other methods are proposed,which are applied to semantic segmentation of street scene.Firstly,this paper expounds the basic theory of semantic segmentation,studies the basic principle of semantic understanding and semantic modeling method,and studies the basic principle of super-pixel segmentation method.Aiming at SLIC super-pixel algorithm,the principle of its application is studied,and several superpixel segmentation methods are designed to verify the performance of SLIC superpixel segmentation methods.The SLIC algorithm is used to design specific experimental methods to preprocess the experimental images in the later stage.Secondly,an image semantics segmentation method based on super-pixel segmentation is proposed,which combines random forest with traditional feature extraction methods.On the basis of super-pixel segmentation,this method uses traditional color,shape,texture and spatial structure information to extract features,and combines random forest classifier to realize segmentation of different objects.In this process,the theory of four feature extraction methods is studied first,and the corresponding experiments are designed to verify the effect of feature extraction.Secondly,the random forest theory is studied,and the design experiment compares the performance of several different classifiers.Finally,the overall experimental scheme is designed.The results show that the segmentation accuracy is 72.4% when the number of superpixels is 300.It is verified that the proposed method achieves better segmentation results.Finally,based on the above research,a semantic segmentation method combining random forest method and deep learning method is proposed.The basic principle of CNN convolutional neural network is studied,and an appropriate feature extraction method is proposed according to its principle.The concrete method of combining CNN feature extraction with stochastic forest algorithm is proposed,and the overall flow chart of the model is designed.Finally,the design is verified according to the proposed scheme.The experimental results were analyzed and compared.The experimental results show that the PA value is 83.2% and the MIoU value is 24.7%.It is verified that the proposed method has good practical results.