基于音频和近红外光谱融合技术的西瓜成熟度判别

Distinguishing watermelon maturity based on acoustic characteristics and near infrared spectroscopy fusion technology

为了满足西瓜成熟度的快速无损检测需求,该研究主要利用声学技术、近红外光谱技术结合K最近邻法(k-nearestneighbor,KNN)、线性判别分析(linear discriminant analysis,LDA)和反向传播人工神经网络(back propagation artificialneural network,BP-ANN)3种化学计量学方法对不同成熟度的西瓜进行定性判别;同时采用联合区间偏最小二乘筛选法(synergy interval partial least squares,Si-PLS)分别建立声学技术、近红外光谱技术、融合技术的西瓜可溶性固形物预测模型。结果表明融合技术处理结果均优于单一信号,其LDA模型数据的西瓜成熟度模型识别率较佳,校正集和预测集的识别率分别为 100.00%和 91.67%。同时,基于融合技术所建立的西瓜可溶性固形物预测模型效果较佳,其校正集的均方差根误差(root mean squared error of the calibration set, RMSECV)为0.601%,预测集的均方差误差(root mean squared errorof the prediction set,RMSEP)为0.725%,相比的单独音频信号其均方根误差分别降低了0.081、0.068个百分点。研究结果可为高精度的西瓜品质快速鉴别提供参考。

Rapid non-destructive detection internal quality of fruit can improve the marketable value of watermelon. Non-destructive technology discrimination and quality assessment of watermelon in real time is a hot topic. More and more studies have been used for watermelon quality detection, such as near infrared spectroscopy, acoustic methods, electrical and magnetic methods etc. Compared with non-destructive methods such as near-infrared spectroscopy and nuclear magnetic resonance, acoustic signal detection has the advantages of simple equipment and low price. So it is widely used to detect the internal quality information of watermelon. Near-infrared (NIR) spectroscopy is a suitable method to characterize organic compounds which have been successfully applied for the detection of soluble solids content (SSC). However, the accuracy of single detection method is not very high. In this study, we aimed to combine acoustic technology with near-infrared spectroscopy of discrimination and watermelon quality at three maturity level. For the acoustic signal of watermelons, a simple single wire pendulum platform was built for the collection of acoustic. Then the acoustic signal was converted into a spectrum signal. The spectrum signal was modeled after extracting feature variable by (Principal Component Analysis) PCA method. Linear discriminant analysis (LDA), K-near neighbor (KNN) and back-propagation artificial neural network (BP-ANN) were successfully discriminated the watermelon based on the maturity stages. Meanwhile, SSC was the most important evaluation index of watermelon. Therefore synergy interval partial least squares (Si-PLS) was used to establish the SSC prediction of watermelon multivariate model of acoustic signal, spectral signal and conjunction information. The LDA of concatenated technique was the best calibration model and the recognition rate of the calibration set and prediction set get highest were 100.00% and 83.33%, respectively. The rc was 0.846 9 and rp was 0.723 3 of Si-PLS model, respectively. For the near-infrared spectroscopy of watermelon, the study used a portable near-infrared spectrometer to collect information about the quality of watermelons at the equator. Then the spectral signal was collected by portable NIR spectrometer for characteristic wavelength screening by GA (Genetic Algorithm). Finally, the 126 variables were obtained to the next modeling analysis. BP-ANN model showed the highest accuracy while Si-PLS model presented higher prediction (rp was 0.834 8) and calibration (rc was 0.855 9). For the acoustic signal merged the near-infrared spectral signature signal, maximum and minimum normalization of the acoustic signal and the spectral signal were obtained. The precision of quantitative and qualitative models were improved. LDA of concatenated technique showed the best calibration model and the recognition rate of the calibration set and prediction set were 100.00% and 91.67%. And the established calibration model had the best effect. The root mean squared error of the calibration set (RMSECV) was 0.601%, and the root mean squared error of the prediction set (RMSEP) was 0.725%. The rc was 0.901 5 and rp was 0.850 6. Compared with the individual audio signals, the root mean square error is reduced by 0.081 and 0.068 percentage points, respectively. Thus, it evidenced that the acoustic signal merged the near-infrared spectral signature signal can improve the accuracy of the qualitative judgment of watermelon maturity. Based on portable NIR spectroscopy and acoustic signal, the rapid prediction of soluble solids in watermelon can be achieved.