Feature Extraction And Classification Of The Multi-Objective Complex Mining Image

Ore type determines the field of its application and practical value, the classification is to establish a mine ore mining prediction model and can be the basis for comprehensive analysis of mineral resources, the on-line ore classification system therefore has very important significance. Intelligent computer-based classification system is more efficient than the traditional manual classification. And it even can be able to provide more accurate classification results. Because of the accumulation of belt ore image, makes it difficult to distinguish the ore and the background. And also because of the poor image acquisition environment, makes the ores'accuracy classification is very difficult.This paper focuses on the feature extraction and classification problem of the Multi-objective complex mining image. The feature extraction is the base of the ore's accurate classification. Therefore, this paper mainly study the multi-objective color,texture and shape feature extraction of the complex image. Because the collected ambiguous image including much noisy, so we first use a series of pre-processing to improve the quantity of the image. based on the watershed segmentation we use the freeman chain code to extract the ore's shape features; and based on the RGB and HSV color space extract the ore region's color feature; after that based on statistical co-occurrence matrix extract the region's texture features. In order to consider the whole features of the shape, color, and texture features, based on the ore regions proposed an integrated feature extraction techniques, the method can be taken into account the overall objectives of ore characteristics and the internal details of the characteristics, By using the support vector machine to class the extracted feature vectors, the result of the classification reflect the characteristics of the ore.This article researches the kmeans cluster of the image pixel color and texture integrated features, the SVM classification of the region pixel color,texture and shape integrated features, and the SVM classification of the region pixel PCA reduction features. The results show that the shape feature is conducive to the proper classification of the ore target, ore recognition rate has been markedly improved; the color, texture and shape integrated reduction features of the ore is also conducive to the proper classification。...