Classification of Codling Moth-Infested Apples Using Sensor Data Fusion of Acoustic and Hyperspectral Features Coupled with Machine Learning

Abstract

Codling moth (CM) is a major apple pest. Current manual method of detection is not very effective. The development of nondestructive monitoring and detection methods has the potential to reduce postharvest losses from CM infestation. Previous work from our group demonstrated the effectiveness of hyperspectral imaging (HSI) and acoustic methods as suitable techniques for nondestructive CM infestation detection and classification in apples. However, both have limitations that can be addressed by the strengths of the other. For example, acoustic methods are incapable of detecting external CM symptoms but can determine internal pest activities and morphological damage, whereas HSI is only capable of detecting the changes and damage to apple surfaces and up to a few mm inward; it cannot detect live CM activity in apples. This study investigated the possibility of sensor data fusion from HSI and acoustic signals to improve the detection of CM infestation in apples. The time and frequency domain acoustic features were combined with the spectral features obtained from the HSI, and various classification models were applied. The results showed that sensor data fusion using selected combined features (mid-level) from the sensor data and three apple varieties gave a high classification rate in terms of performance and reduced the model complexity with an accuracy up to 94% using the AdaBoost classifier, when only six acoustic and six HSI features were applied. This result affirms that the sensor fusion technique can improve CM infestation detection in pome fruits such as apples.