A novel quantitative approach of multidimensional interactive sensing based on information entropy was developed for the rapid determination of rice quality. Electronic tongue with multi-metal sensor array was employed. Physicochemical indexes including chalkiness, gel consistency, amylose, protein, starch and total metal element content which are the major indicators for rice quality were analyzed. Wavelet packet decomposition and fast Fourier transform were used for the decomposition and transformation of the original voltammetric signal. The square color block diagram and the dial color block diagram were used for the characterization. The multidimensional interaction matrix was constructed by information entropy. CNN model, BpNN model and the federated model (CNN + BpNN) were established to the quantitative prediction for the physicochemical indexes of rice. Compared with CNN and BpNN model, the accuracies of CNN + BpNN model were the highest. The training accuracies and prediction accuracies of CNN + BpNN with MMxI-3 as the input for all physicochemical indexes were 84.3%～92.0% and 81.9%～89.5% respectively, which were higher than those of other multidimensional interaction matrices as well as the original characteristic matrix as the input. Results indicated that the multidimensional interaction matrix contained more quantitative information in the sensor array for physicochemical components. In conclusion, the combination of the federated model and multidimensional interaction matrix for electronic tongue sensor array could be used as an effective approach for the quantification of rice quality.

提出了一种基于信息熵的多维交互式传感定量方法，用于水稻品质的快速测定。使用带有多金属传感器阵列的电子舌。分析了稻米品质的主要指标，包括垩白度，凝胶稠度，直链淀粉，蛋白质，淀粉和总金属元素含量的理化指标。小波包分解和快速傅里叶变换被用于原始伏安信号的分解和变换。方形方框图和表盘方框图用于表征。多维交互矩阵是通过信息熵构造的。CNN模型，建立了BpNN模型和联邦模型（CNN + BpNN），对水稻的理化指标进行了定量预测。与CNN和BpNN模型相比，CNN + BpNN模型的准确性最高。以MMxI-3作为输入值的CNN + BpNN的训练精度和预测精度分别为84.3％〜92.0％和81.9％〜89.5％，高于其他多维交互矩阵和原始矩阵特征矩阵作为输入。结果表明，多维相互作用矩阵在物理化学成分传感器阵列中包含更多定量信息。结论，