基于特征融合的低倍镜下大鼠骨质疏松识别

发布时间：2018-10-15 19:40

【摘要】：低倍镜下纹理特征不清晰,并且相较于高倍镜更易受到方向和距离的影响。为此我们提出改进的纹理特征计算方法,结合形状分析方法,进行特征级融合以提高骨质疏松识别准确率。通过对纹理参数在方向和距离上的分析,发现相关和短游程矩对方向非常敏感,并且距离在等于3时各参数变化趋于稳定。基于变异系数法利用相关和短游程矩计算各个方向权重系数,使用距离等于3时的纹理数据得到最终纹理参数结果,融合形状参数,用线性支持向量机、K-最近邻分类算法和线性判别分析方法进行分类识别。采用加权纹理参数比常规未加权的纹理参数识别准确率高,同时融合了形状参数后识别准确率比仅用纹理参数高。线性判别分析识别率最高达到了92.3%。采用加权纹理融合形状参数的方法识别准确率显著提高,具有诊断应用价值。
[Abstract]:The texture feature of low power mirror is not clear, and it is more easily affected by direction and distance than high power mirror. In order to improve the accuracy of osteoporosis recognition, we proposed an improved texture feature calculation method, combined with shape analysis method, to perform feature level fusion. By analyzing the direction and distance of texture parameters, it is found that correlation and short run moment are very sensitive to each other, and the variation of each parameter is stable when the distance is equal to 3. Based on the variation coefficient method, correlation and short run-time moments are used to calculate the weight coefficients in each direction. The final texture parameters are obtained by using the texture data with distance equal to 3, and the shape parameters are fused. Linear support vector machine (LSVM), K-nearest neighbor classification algorithm and linear discriminant analysis (LDA) are used for classification and recognition. The recognition accuracy of weighted texture parameters is higher than that of conventional unweighted texture parameters, and the recognition accuracy is higher than that of only texture parameters after the shape parameters are fused. The highest recognition rate of linear discriminant analysis is 92.3%. The recognition accuracy of shape parameters by weighted texture fusion is improved significantly, which is valuable for diagnosis.
【作者单位】：广东医科大学信息工程学院;广东医科大学基础医学院;
【基金】：国家自然科学基金资助项目(81541104) 广东省科技计划项目(2014A030304006) 湛江市科技专项竞争性分配项目(2015A01039) 广东医科大学科研基金项目(M2016029)
【分类号】：R580;TP391.41

【参考文献】