具有自学习和邻域搜索能力的改进蚱蜢优化算法及红外图像分割应用

IMPROVED GRASSHOPPER OPTIMIZATION ALGORITHM WITH SELF-LEARNING AND NEIGHBORHOOD SEARCHING ABILITY AND ITS APPLICATION ON INFRARED IMAGE SEGMENTATION

  • 摘要: 传统图像分割方法处理电力设备红外图像分割问题时存在精度低、诊断效率差的不足。提出一种具有自学习和邻域搜索能力的改进蚱蜢优化算法IGOA,并结合Cross熵应用于电力设备红外图像分割。为了提升标准蚱蜢优化算法GOA的寻优精度和寻优速率,利用佳点集、伪对立学习、配对自学习及邻域搜索策略对GOA的全局寻优能力进行改进。然后以Cross熵作为评估标准,构建红外图像分割模型IGOA-Cross。利用四种常规电力设备红外图像进行实验分析,结果表明:与对比模型相比,该分割模型误分率更低,峰值信噪比和结构相似度更高,能够处理背景非均匀、噪声较大的红外图像分割,分割效率和精度都有提升。

     

    Abstract: When deal with infrared image segmentation of electrical equipment, traditional image segmentation method has the shortage of the low accuracy and poor efficiency of diagnosis. Therefore, a improved grasshopper optimization algorithm (IGOA) with self-learning and neighborhood searching ability is proposed. Combined with cross entropy, the improved algorithm was applied on infrared image segmentation of electrical equipment. In order to improve the optimization precision and the optimizing efficiency of standard grasshopper optimization algorithm (GOA), the good point-set, pseudo-opposite learning, paired self-learning and neighborhood searching strategy were used to improve the global optimizing ability of GOA. JP3Cross entropy was regarded as the evaluation measurement, a segmentation model of infrared images IGOA-Cross was constructed. Four common infrared image of electrical equipment were used for experimentalJP analysis. The results show that compared with contrast model, the proposed model has lower misclassification rate, higher peak signal-to-noise ratio and structural similarity degree. The improved model can deal with infrared image segmentation with inhomogeneous background and large noise, and has a better segmentation efficiency and precision.

     

/

返回文章
返回