Image Semantic Segmentation Scheme based on XGBoost combination with Convolution Feature Extraction

Aiming at the problem that the existing image semantic segmentation schemes based on convolutional neural network require a large number of training data sets and high training cost, this paper presents a novel scheme for image semantic segmentation based on extreme gradient boosting (XGBoost), a scalable tree boosting system and the combination of convolution feature extraction (CFE). The scheme consists of three stages. First, convolution operation is performed on a small amount of RGB pixel information of the image processed by Gaussian filtering to acquire more multi-dimensional image feature information. Then, some preprocessing techniques of image feature information, including feature normalization and image label pixel number uniformity are performed to match the model. Finally, the preprocessed image feature information and corresponding labels are input into XGBoost to train and predict the data of the test set. Through simulation and practical tests, compared with other commonly used few data classification algorithms like k-Nearest-Neighbor (kNN) and Support-Vector-Machine (SVM), this scheme can achieve high precision pixel-level image semantic segmentation under the condition of lower training cost.