极限学习机集成在骨髓细胞分类中的应用

骨髓细胞的分类有重要的医学诊断意义。先对骨髓细胞图像分割和特征提取，用提取出来的训练集对极限学习机训练，再用该分类器对未知样本识别。针对单个分类器性能的不稳定，提出基于元胞自动机的极限学习机集成算法。通过元胞自动机抽样策略构建差异大的训练子集，多个分类器并行学习，多数投票法联合决策。实验结果表明，与BP、支持向量机比较，该算法基本无参数调整，学习速度快，分类精度高能达到97.33%，且有效克服了神经网络分类器不稳定的缺点。

Classification of bone marrow cells based on ensemble of extreme learning machine

Classification of bone marrow cells has important medical diagnostic significance. The training samples set extracted from the segmented images of bone marrow cells is used to train the extreme learning machine. Then this trained extreme learning machine automatically classifies the unknown bone marrow cells. For the instability of performance of single classifier, the ensemble of extreme learning machine algorithm based on cellular automata is proposed. The different training subsets are constructed by cellular automata strategy through sampling, then they are learned in parallel with multiple classifiers, finally the outputs are combined by majority voting. Experimental results show that this proposed algorithm has fast learning speed and gains high classification accuracy reached 97.33% without adjusting any parameters during run-time compared with BP neural networks and support vector machines. Moreover, it effectively solves the disadvantage of instability for the neural network classifier.