宽度学习系统在蘑菇毒性判别中的应用

Application of Broad Learning System in Discrimination of Mushroom Toxicity

摘要:

为了提升蘑菇毒性判别的准确率，消除个体差异，本文提出了一种基于宽度学习系统的蘑菇毒性判别方法。文章首先对蘑菇各特征指标与其毒性判别的相关性进行了探究，其所得结果显示蘑菇的气味和颜色是其区分度最大的特征，该结果与人工判别积累的经验相符。接着构建宽度学习系统并进行训练，对比分析不同样本量情况下所提方法的准确率，发现当样本数量大于1000时，宽度学习系统分类准确率便高于99.5%。与BP-神经网络相比，该方法准确率高且所需训练时间短。最后依据宽度学习系统的增量学习算法，在模型性能不满足要求时，通过增加隐藏层节点快速更新系统，使系统分类准确率从98.55%提升至99.99%，而不需要将网络重新进行训练，这使实时判别蘑菇毒性成为可能。因此对比其他方法，基于宽度学习系统的蘑菇毒性判别方法具有准确率高、训练时间短、判别迅速且易拓展的优点。

Abstract:

In order to improve the accuracy of mushroom toxicity discrimination and eliminate individual differences, a method of mushroom toxicity discrimination based on broad learning system was proposed in this work. Firstly, the correlation between each characteristic of mushroom and its toxicity was explored. The results showed that the odor and color of mushroom were the most distinguishing characteristics. These results were consistent with the experience of manual discrimination. Then, broad learning system was established and trained. By comparing performance in diverse sample sizes of different methods, it was found that when the sample size is larger than 1000, the classification accuracy of the broad learning system was higher than 99.5%. Compared with BP-neural network, the proposed method was of high-accuracy and fast-training. Finally, according to the incremental learning algorithm of broad learning system, when the performance of the system does not meet the requirements, the system can be updated quickly by increasing the hidden nodes, and the accuracy can be improved from 98.55% to 99.99% without retraining the whole network. It was possible to discriminate the toxicity of mushrooms in real time. Therefore, compared with other methods, this method of mushroom toxicity discrimination based on broad learning system has the advantages of high accuracy, short training time, rapid discrimination and easy expansion.