The goal of this paper was to classify blood cells based on hyperspectral data range from near infrared to visible light. Different from common classification method, the extracted features included not only gray information in flat images, but also plentiful spectral characteristics. As for the blood cell classification, the classifier was made up of genetic algorithm and neural network. Experimental results show that this means is effective on red cells and nucleolus of tumor cells. Compare to spectral data of 10 bands and data of 80 bands, hyperspectra can get better results at the cost of running time.%对近红外—可见光范围内的超高光谱血液图像进行血细胞分类.不同于常见的血细胞识别方法,对血细胞的特征提取不但含有图像灰度特征,而且还包含了丰富的光谱特征,在分类方法上利用具有自适应能力的遗传算法和神经网络设计分类器进行血细胞的分类.实验结果表明,该法对背景点、红细胞和病变细胞核可取得比较好的识别结果.相对于10波段光谱和80波段高光谱而言,220波段的超光谱以增加运行时间为代价,取得了较好的分类结果.
展开▼
