Enhanced Insect Classification: A Comparative Study Using CNN Models

Valid species identification is of paramount importance in biodiversity research and ecological monitoring. This research explores the performance of convolutional neural networks (CNNs) in identifying butterfly and moth species, evaluates one-against-all as well as wholly connected features of transfer learning pre-trained models with a 100 mixed species data set including images taken under different conditions for each separated class. We chose to experiment with five models, VGG16, VGG19, ResNet152, InceptionV3, and DenseNet201 architecture, for the simple fact that they are proven among the best when it comes to categorizing images in general. We fine-tuned both models and evaluated them using pre-extracted features from the ImageNet dataset. The objective of this Study is to compare all models in order to get the best result for recognition accuracy and generalization from butterfly moth species.