Maturity Grading of Cucumber Based on Hyperspectral Imaging and Deep Learning

The maturity of cucumbers within the same growth period can vary due to environmental factors, affecting their storage quality and eating experience. Accurate postharvest maturity grading of cucumbers can increase their commercial value and reduce waste. However, current methods for determining cucumber maturity rely primarily on the subjective experience of harvesters and the number of days after flowering. This approach is highly subjective, labor-intensive, and increasingly costly. To address the aforementioned issues, this study used hyperspectral images of cucumbers at different maturity grades as the data source and proposed a novel cucumber maturity grading method. In the first stage, spatial–spectral preprocessing and feature compression are performed on hyperspectral images to generate a compact feature matrix. In the second stage, the feature matrix is fed into a Transformer-CNN-BiLSTM (TCBL) model for maturity grading. Experimental results show that when compared with traditional machine learning methods based on band selection and typical deep learning models, the proposed method achieved the best performance in maturity grading tasks, with accuracy, precision, F1-score, and recall rates all exceeding 98%, representing a 5% improvement in grading accuracy over traditional methods. This method provides theoretical and quantitative data support for the development of online cucumber maturity grading systems.