Soft-Shell Prawns (Penaeus monodon) Can Be Identified Using Hyperspectral Imaging and Machine Learning – A Novel Approach to Ensure Consistency, Accuracy and Speed

Abstract Soft-shell prawns (shrimps), which yield lower market value than their hard-shell counterparts, are currently identified through manual inspection—a subjective and inconsistent process. This study explores the use of shortwave infrared (SWIR; 950–2515 nm) hyperspectral imaging (HSI) combined with machine learning as a real-time, non-destructive alternative for classifying prawn shell hardness. A total of 380 farmed prawns spanning four manually assessed hardness categories were scanned using a HSI camera. Two classification models—support vector machine (SVM) and partial least squares discriminant analysis (PLS-DA)—were trained on 50% of the total samples (training data set) to associate spectral signatures with shell hardness classes and evaluated using the remaining 50% (independent test data set). PLS-DA marginally outperformed SVM in overall classification accuracy, achieving 92.1%, compared to SVM’s 90% on the independent test set. Although SVM showed better performance for intermediate hardness classes, its higher misclassification rate for the extreme classes (hard and soft) made it slightly less reliable for practical application. Reduced sensitivity in the intermediate classes across both models likely stems from limited sample size and subjectivity in the manual reference classifications. These results demonstrate the potential of HSI as a consistent and objective tool for prawn classification, offering significant advantages for automating shell hardness assessment and sorting. Implementing this technology could enhance processing efficiency and product quality within the prawn industry.