Efficient durian sugar content grading via hyperspectral imaging and fast continuous wavelet transform and spiking neural network framework

To achieve efficient and rapid detection of durian sugar content—with potential extension to maturity assessment of other fruits—this study proposes a novel hybrid framework integrating Fast Continuous Wavelet Transform (FCWT) and Spatiotemporal Backpropagation-based Spiking Neural Network (STBP-SNN). Hyperspectral images of durian pulp were collected within the 900–1700 nm wavelength range, and samples were categorized into three sugar levels (high, medium, low) to establish a targeted evaluation system for sugar content grading.​FCWT was employed for multi-scale time-frequency analysis, converting one-dimensional spectral curves into two-dimensional feature matrices to capture fine-grained spectral variations that are critical for distinguishing sugar levels. These feature matrices were then input to the STBP-SNN, which leverages the unique dynamics of spiking neurons to effectively extract spatiotemporal features from hyperspectral data and perform accurate classification.​ The experimental results indicated that the proposed framework achieved an average test accuracy of around 96%, highlighting its robustness in discriminating durian sugar content. By integrating advanced spectral feature extraction (FCWT) with intelligent spatiotemporal classification (STBP-SNN), the framework provides a scalable and reliable solution for fruit quality assessment, especially for precision grading of internal quality indicators like sugar content.