Avocado fruit sorting by hyperspectral images

Abstract

The paper shows the use of the methods of hyperspectral imaging (HSI) in a range of 400–1000 nm and multivariate analysis for sorting Hass avocado fruits. The decomposition of the data matrix of HSIs of avocado fruits was carried out using the principle component analysis. The reflection bands in the visible and near-infrared spectral regions interrelated with the process of maturation and the moisture content of avocado fruits were revealed. It has been established that visualization upon avocado inline sorting by moisture is possible when using factor loadings as pseudo-color. Calibration models for determination of moisture and dry matter of avocado fruits were built based on the data of moisture measurement and hyperspectral images. The matrix of spectral data was formed by two methods: random selection of spectral signatures of HSIs from the whole surface of fruits or the image surface of HSIs of fruits (initial HSIs) as a region of interest (ROI). Based on the data of moisture measurement and selection of spectral signatures of hyperspectral images, calibration models were built for detection of moisture and dry matter of avocado fruits. Using sequential simulation by the projection to latent structures (PLS) method, accurate calibration models were developed to detect moisture (RP2 = 0.89) and dry matter (RP2 = 0.92) in the composition of avocado fruits. When building calibration models by the initial HSIs, models were obtained to predict moisture (RС2 = 0.99) and dry matter (RС2 = 0.99) in the composition of avocado fruits. It is proposed to use calibration models by the initial HSIs to determine moisture and dry matter in the intervals of the acceptable values according to the acting standard UNECE STANDARD FFV-42:2019.

Keywords

• avocado
• quality
• hyperspectral imaging
• multivariate analysis
• moisture
• dry matter
• pls