Multi-sensor data fusion for improved prediction of apple fruit firmness and soluble solids content

摘要

Several nondestructive technologies have been developed for assessing the firmness and soluble solids content (SSC) of apples. Each of these technologies has its merits and limitations in predicting the two quality parameters. With the concept of multi-sensor data fusion, different sensors would work synergistically and complementarily to improve the quality prediction of apples. In this research, four sensing systems (i.e., an acoustic sensor, a bioyield firmness tester, a miniature near-infrared (NIR) spectrometer, and an online hyperspectral scattering system) were evaluated and combined for nondestructive prediction of firmness and SSC of 'Jonagold' (JG), 'Golden Delicious' (GD), and 'Delicious' (RD) apples. A total of 6,535 apples harvested in 2009 and 2010 were used for analysis. Each of the four sensors showed various degrees of ability to predict apple quality. Better predictions of the firmness and, in most cases, of the SSC were obtained using sensors fusion than using individual sensors, as measured by number of latent variables, correlation coefficient, and standard error of prediction (SEP). Results obtained from the two harvest seasons with the multi-sensor fusion approach were quite consistent, confirming the validity and robustness of the proposed approach. The SEPs for firmness measurement of JG, GD and RD using the best combination of two-sensor data were reduced by 13.3, 19.7 and 7.9% for the 2009 data and 16.0, 12.6 and 4.7% for the 2010 data; and using all four-sensor data by 21.8, 25.6 and 13.6% in 2009, and 14.9, 21.9, and 7.9% in 2010, respectively. For SSC prediction, using the two-sensor data (i.e., NIR and scattering) improved predictions for JG, GD and RD apples harvested in 2009, with their SEP values being reduced by 10.4, 6.6 and 6.8%, respectively. This research demonstrated that the fused systems provided more complete complementary information and, thus, were more powerful than individual sensors in prediction of apple quality.