The Effect of Data Augmentation and Optimization Technique on the Performance of EfficientNetV2 for Plant-Parasitic Nematode Identification

Abstract

Plant-parasitic nematodes are major agricultural pathogens contributing to massive crop losses worldwide. It is crucial to identify plant-parasitic nematodes to decide the best pest control and management strategy. The current identification technique is based on visual observation from nematode microscopic images done by the nematologist. However, this method requires a long process and is prone to error. A deep learning-based method can be implemented to speed up the current identification process. This study explores the effect of combining several data augmentation techniques, namely brightness, contrast, blur, and noise, on the performance of the EfficientNetV2B0 and EfficientNetV2M models for identifying plant-parasitic nematodes. Moreover, this study also compared three optimizers while training the models to find the best optimizer for each model and data augmentation. The results show that the EfficientNetV2B0 model yielded the highest test accuracy of 96.91% when employing no augmentation and trained using SGD and RMSProp optimizer. Furthermore, the EfficientNetV2M model gave the highest test accuracy of 96.91% when the combination of brightness and contrast augmentations was applied and trained using the RMSProp optimizer.