AUTHOR=Zhang Chao , Shi Yue , Wei Zhonghui , Wang Ruiqi , Li Ting , Wang Yubin , Zhao Xiaoyan , Gu Xiaohe TITLE=Hyperspectral estimation of the soluble solid content of intact netted melons decomposed by continuous wavelet transform JOURNAL=Frontiers in Physics VOLUME=10 YEAR=2022 URL=https://www.frontiersin.org/journals/physics/articles/10.3389/fphy.2022.1034982 DOI=10.3389/fphy.2022.1034982 ISSN=2296-424X ABSTRACT=

Netted melons are welcomed for their soft and sweet pulp and strong aroma during the best-tasting period. The best-tasting period was highly correlated with its soluble solid content (SSC). However, the SSC of the intact melon was difficult to determine due to the low relationship between the hardness, color, or appearance of fruit peel and its SSC. Consequently, a rapid, accurate, and non-destructive method to determine the SSC of netted melons was the key to determining the best-tasting period. A hyperspectral model was constructed to estimate the SSC of intact netted melons. The combination of continuous wavelet transform and partial least squares or random forest algorithm was employed to improve the estimation accuracy of the hyperspectral model. Specifically, the hyperspectra of the diffuse reflection and SSC of 261 fruit samples were collected. The sensitivity band was screened based on the correlation analysis and continuous wavelet transform decomposition. The correlation coefficient and RMSE of the random forest regression model decomposed by the continuous wavelet transform were 0.72 and 0.98%, respectively. The decomposition of the continuous wavelet transform improved the correlation coefficient by 5 and 1.178 times at 754 and 880 nm, respectively. The random forest regression model enhanced the determination coefficient by at least 56.5% than the partial least squares regression model, and the continuous wavelet transform decomposition further enhanced the determination coefficient of the random forest regression model by 4.34%. Meanwhile, the RMSE of the random forest regression model was reduced. Therefore, the decomposition of the continuous wavelet transform improved the stability and prediction ability of the random forest regression model.