林火烟雾图像自动识别中的模式分类器选择

探索了支持向量机（SVM）方法解决由脉冲耦合神经网络（PCNN）提取的林火烟雾图像特征后的计算机视觉模式识别问题。针对由于林火烟雾图像的纹理特征不突出，即便用特殊方法提取出来的特征向量也维数较高，对后续分类器性能提出较高要求并且分类效果存在很大的未知性等问题，通过实验，对3种人工神经网络分类器和支持向量机分类器的烟雾图像特征甄别效果进行了详细对比。结果表明:基于支持向量机的分类器在复杂的森林背景情况下对烟雾有很好的分辨能力，其识别准确率达到94.26%，并且在识别准确率和分类速度两方面都超过了作为对照的3种神经网络分类器。 

Selection of pattern classifier in automatic detection for forest fire smoke feature.

The performance of pattern recognition methods for forest fire smoke feature extracted by pulse-coupled neural network(PCNN) was explored in this paper.PCNN smoke feature require high performance to classifiers due to the relevantly high dimension of extracted eigenvectors and the vague nature of smoke.These problems also bring uncertainty to the recognition.Comparative experiments between artificial neural networks(ANNs) and SVM were proposed.Experimental results show that SVM method outperforms other classifiers and the accuracy reaches 94.26% based on our smoke image database.