基于改进的Cifar-10深度学习模型的金钱豹个体识别研究

Individual Identification of Leopard Based on Improved Cifar-10 Deep Learning Model

将深度学习方法首次应用到金钱豹个体识别研究中,通过对山西沃成生态环境研究所2010-2016年的金钱豹影像数据进行清洗及预处理,得到训练图片540张,测试图片110张。基于Cifar-10深度学习模型,使用"Dropout"防止模型训练结果过拟合,采用多种池化方式组合训练,得到较优的金钱豹个体识别深度学习模型。最终得到的深度学习模型的识别准确率可以达到99.3%,能够有效地识别金钱豹个体。

With the decrease of leopard population,the state and society attach importance to the conservation of leopard.Individual identification of leopard is the basis for effective conservation of leopard populations.In this paper,deep learning was first applied to leopard identification.The leopard images from Shanxi Wocheng Institute of Ecological and Environment in 2010-2016 were given data cleaning and data preprocessing,getting 540 training pictures and 110 test pictures.On the basis of the Cifar-10 deep learning model,the"Dropout"model was used to prevent the training results from overfitting,the pooling methods were used for combined training to get a better deep learning model for individual identification of leopard.This deep learning model has the accuracy of 99.33%in individual identification of leopard,and can effectively realize the individual identification of leopard.