Internal damage inspection of almond nuts using optimal near-infrared waveband selection technique

• An advanced waveband selection method for near-infrared spectral data is presented.
• Only four wavebands are needed for quality inspection of almond nuts.
• Our proposed method yields a higher classification rate than prior methods.
• Our classification model is suitable for commercial online processing.

This work presents a statistical method for internal damage inspection of almond nuts based on advanced waveband selection and supervised pattern recognition techniques using near-infrared spectral data. Our proposed method employs an optimal adaptive branch and bound algorithm to select a small set of wavebands for use in a support vector machine classifier. Our case study involves discriminating almond nuts with internal damage from normal ones. Experimental results demonstrate that our method gives significantly higher classification rates than prior algorithms. Our classification model is promising for commercial online processing, since only a few wavebands are used for classification and can thus be recorded by many fast sensor systems.