Mango Fruit Ripening Degree Assessment Based on Convolutional Neural Network and Hyperspectral Imaging Technology

Abstract:

In order to non-destructively determine mango ripeness, this study focused on exploring the combined use of visible near-infrared hyperspectral imaging and deep learning methods. The hyperspectral data of unripe, ripe and overripe mangoes were collected in the wavelength range of 400~1 000 nm using a diffuse reflectance hyperspectral imaging system. The SPXY algorithm was used to divide the data into training, validation and testing sets in a ratio ofo 6:2:2. After the spectral pre-processing such as Wavelet Threshold Denoising (WTD) and Multiplicative Scatter Correction (MSC), a 1-Dimensional Convolutional Neural Network (1DCNN) model was constructed. Meanwhile, Principal Component Analysis (PCA) and Bootstrapping Soft Shrinkage (BOSS) were used to extract the feature bands, and the classification models of Partial Least Squares-Discriminant Analysis (PLS-DS) and Extreme learning machine (ELM) were constructed. Comparative analysis shows that the 1DCNN model based on multiple scatter correction preprocessing performed the best, with a classification accuracy of 98.30% and a Kappa coefficient of 0.97 on the test set. The study concludes that the combination of hyperspectral imaging technology and neural network algorithm can reliably and accurately determine mango ripeness, providing a theoretical basis for the practical application of automated mango sorting.