Plant Parasitic Nematodes Classification using Pre-trained CNN Models: ResNet50 and VGG16

Abstract

Parasitic nematodes cause serious crop damage worldwide, resulting in significant financial losses. Approximately 0.01 % of these species have not yet been found, according to estimates from certain experts. Nematodes can be challenging to classify using conventional methods because most of them share similar morphological characteristics. In the past, nematodes could only be distinguished by their morphological qualities, which include body length, the orientation of their reproductive organs, and other physical attributes. The previously mentioned approach requires a great deal of labor and expertise, and it only uses expensive machinery and human skills to classify objects. DL-based methods have significantly improved and increased accuracy in recent years. These species were successfully classified using the DL algorithms ResNet50 and VGG16. Acrobeles, Acrobeloides, Aphelenchoides, Amplimerlinius, and Discolimus are the five species of nematodes that were employed. The provided dataset, which originally contained 1500 digital images of nematodes, is further increased to 5000 images by the use of data augmentation techniques such as zooming, flipping, shearing, and other processes. ResNet50 and VGG16, two pre-trained CNN models, have been enhanced to better categorize these species. The accuracy rates of the VGG16 and ResNet50 models are 95.87 % and 98.02 %, respectively.