基于小波变换的乳酪制品智能鉴别技术

实验采集了不同品牌乳酪制品的拉曼光谱数据,经过不同小波变换处理后输入k近邻算法,建立了一个新型的优化识别流程。结果显示,未经任何谱图处理时,鉴别算法的识别准确率仅为82.27%,经小波软阈值去除信号噪声影响(coif1小波基,分解尺度n=4)后识别率可提升至86%,经sym5小波增强后,识别率可达87.07%,再经融合处理,识别率可进一步达到88.6%。在此基础上进一步研究了归一化结合小波变换处理优化鉴别算法的情况,发现经小波降噪及归一化处理至0,1]区间后,k近邻鉴别算法识别率可达93.73%,再经sym5小波增强后,识别率可达94.4%,最后经融合处理,识别率可达到95.4%。拉曼光谱数据经归一化、小波降噪、增强、融合处理后,可有效提高鉴别算法的识别率。

Optimization of Intelligent Identification Technology for Dairy Products Based on Wavelet Transform

It has the advantages of fast calculation speed and objective evaluation.The Raman spectral data of different brands of cheese products were collected in this experiment.After different wavelet transform processing,the data were input to the k-nearest neighbor algorithm,and then a new optimized recognition approach was established.The results show that the recognition accuracy of the identification algorithm is only 82.27%without any spectral processing.After removing the influence of signal noise(coif1 wavelet basis,decomposition scale n=4)by wavelet soft threshold,the recognition rate can be improved to 86%.After sym5 wavelet enhancement,the recognition rate can reach 87.07%.After fusion processing,the recognition rate can further reach 88.6%.It is found that after wavelet denoising and normalization processing to0,1]interval,the recognition rate of k-nearest neighbor algorithm can reach 93.73%.After sym5 wavelet enhancement,the recognition rate can reach 94.4%.Finally,after fusion processing,the recognition rate can reach 95.4%.According to the above analysis,after normalization,wavelet denoising,enhancement and fusion processing,the recognition rate of the identification algorithm can be effectively improved,which provides a technical reference for the quality control of dairy products.