# Soluble solids content prediction of pineapple based on visible-near infrared hyperspectral image

## TL;DR

This study uses hyperspectral imaging to accurately predict the sugar content of pineapples, which is important for quality assessment and market value.

## Contribution

The study introduces a novel hyperspectral imaging method with optimized models for precise soluble solids content prediction in pineapples.

## Key findings

- PLSR model achieved high accuracy with R²=0.9459 and RMSE=0.5746 for SSC prediction.
- The ddA-PLSR model after band selection showed exceptional performance with R²=0.9869 and RMSE=0.1250.
- Four key wavelength ranges were identified as critical for accurate SSC detection.

## Abstract

Pineapple is widely favored by consumers for its rich proteins, vitamin C and other nutrients. Soluble solids content (SSC) has long been the core indicator for pineapple quality assessment, directly affecting its market acceptability and sales. To accurately detect pineapple SSC, this study used a hyperspectral imaging system to collect hyperspectral images in the 400–1700 nm range, with SSC measured by an Atago PAL-1 digital sugar meter as the reference. Five pretreatments (including multiple scattering correction (MSC), polynomial smoothing (SG) and mathematical transformations) were applied to raw spectral data, and three prediction models (partial least squares regression (PLSR), Lasso regression, ridge regression (RR)) were established. All models performed well: PLSR showed R²=0.9459 and RMSE = 0.5746, Lasso R²=0.8965 and RMSE = 1.0221, RR R²=0.8560 and RMSE = 1.2632. After screening characteristic bands via Successive Projections Algorithm (SPA) and re-modeling, the ddA-PLSR model was optimal (R²=0.9869, RMSE = 0.1250), with four key wavelengths (673-676nm, 711-715nm, 971-990nm, 1357-1367nm) extracted. This confirms hyperspectral imaging (HSI) enables efficient and accurate SSC detection in pineapples, with great application potential in pineapple quality identification.