Detection of invisible defects in apple using active thermal imaging with time-series temperature features

Detecting the invisible defects of apples can minimize the loss and control the defect spread, but the existing methods struggle to detect these defects. Considering various defects possess different temperature characteristics, a novel approach using active thermal imaging (TI) and time-series temperature features is proposed. It utilized time-series temperature features with high accuracy. The thermal images were captured by a TI system to form an independent training-prediction split during dual temperature sequences, and time-series temperature features were extracted. Different defects in apples own diverse time-series temperature features. Visible defects show the greatest temperature variation range and the lowest mean temperature, and invisible defects possess the highest mean temperature. Then, the obtained features were combined with support vector machines (SVM) and random forest (RF) to develop a classification model of apple defects. The best determination of defects was obtained by the time-series temperature difference features and SVM with the accuracy of the calibration set and prediction set of 100% and 93.7%. The method proposes an accurate detection of invisible defects in apples and is expected to evaluate the quality of other agricultural products. • Active thermal imaging was used to detect invisible defects in apples. • Thermal images were captured through the dual temperature sequence. • Time-series temperature features were extracted to enhance representation. • Machine learning methods were utilized to develop classification models.