【摘要】 结合卷积降噪自编码器与随机森林算法，提出一种新型的卷积降噪自编码器-随机森林（CDAE-RF）模型，并基于可见-近红外光谱数据集来识别苹果树种。首先，通过网格式搜索、平行实验的方法优化了L1范数等参数，提高了模型的鲁棒性；然后，对比实验分析了CDAE-RF、主成分分析-随机森林模型（PCA-RF）、K最近邻分类算法等方法在不同噪声水平下光谱识别的准确性和鲁棒性。实验结果表明，相对于传统算法，新提出的CDAE-RF模型识别准确率达97.92%，在加噪情况下具有更高的鲁棒性。CDAE-RF模型降低了随机森林算法对噪声的敏感性，提高了噪声光谱图像识别的准确性，为地物波谱识别提供了一种新的方法。更多还原

【Abstract】 Combining the convolutional denoising autoencoder and random forest algorithm, a new convolutional denoising autoencoder-random forest(CDAE-RF) model is proposed to identify apple varieties based on the VIS-NIR spectrum data. Firstly,the L1 norm and other parameters are optimized through grid search or parallel experiments in order to improve the robustness of the model; then, under different noise level, the accuracy and robustness of the proposed CDAE-RF model, principal component analysis-random forest model(PCA-RF) and K-nearest neighbor classification algorithm are analyzed by comparative experiments.Experimental results show that compared with traditional algorithm, the accuracy of the proposed CDAE-RF model is as high as97.92%, and has higher robustness when noise increases. The CDAE-RF model reduces the sensitivity of random forests algorithm to noise, improves the accuracy of noise spectral identification, and provides a new method for feature spectral identification.更多还原