Research On Shadow Detection Of Soil Color Image Based On Semi-supervised

Soil core fracture contains complete characteristics of soil properties,which is often used to identify soil species.The images of soil core fracture are also used as the information carriers of machine vision to identify soil species.Due to the influence of illumination angle and shelter and uneven surface of soil fracture,the images of soil core fracture are collected by machine vision that contain shadow,which may interfere with the later soil identification.Therefore,shadow detection is a necessary preprocessing for soil identification by machine vision.The main purpose of this paper is to detect the shadow of the soil core fracture images collected by machine vision.The main works are as follow:(1)Through the analysis of shadow and non-shadow in HSI color space of soil color images,it is found that shadow and non-shadow have certain separation characteristics in H and I components respectively.M-measure is constructed by using H and I components to further increase the separation characteristics for shadow detection of soil images.The first main peak is obtained by the bimodal characteristic of m-measure histogram,and the second main peak is found out by the relationship between the first main peak and the mean value of m-measure.The segmenting shadow threshold is estimated with two main peaks and part of the data is labelled for shadow and non-shadow.A subset of uncalibrated data is constructed to detect shadow.The discrete degree is constructed by using the subset of data to be detected and the supervision information.Semi-supervised clustering of uncalibrated data is carried out step by step to complete shadow detection of soil color images.The experimental results show the average standard deviations of non-shadow and shadow are 0.063 and 0.058,and the time cost is 0.355s and this work is effective.(2)It is appeared that the segmentation effect of the above method is not accurate for the soil color images with soil chroma and shadow chroma which are similar,while the experimental results of the above method were further analyzed.The characteristics of shadow and non-shadow data in Lab color space of soil color images are studied and the second separability measure ψ is constructed by using L component,a component and b component.ψ/and M that is defined in the above method constitute double separability measures.The method in(1)is used to gain the bimodal value of the double-separability measure.Taking bimodal as the center and r as the radius,the decision value is calculated for the data in the range,and the maximum decision value is extracted step by step.The supervision information is orderly expanded and soil shadows are detected by dispersion that is redefined.The improved shadow detection algorithm based on semi-supervised and double-separability measures is tested with 18 groups of image samples.Experimental results show that the mean standard deviations of non-shadow and shadow are 0.058 and 0.051,respectively,and the average standard deviations are decreased by 0.005 and 0.007,and the time cost is 0.367s.The algorithm improves the accuracy of shadow detection in soil color images.