Identifying an organism requires taxonomic expertise, time, and often adult specimens of that organism. Accurate identification of organisms is of great importance for sustainable agriculture, forestry and fisheries, combating pests and human diseases, disaster management, sustainable trade of biological products and management of alien invasive species. Advances in machine learning techniques have paved the way for the identification of animals by image analysis. In this context, it is aimed to test the success of different convolutional neural network (CNN) models in classifying leaf beetle (Coleoptera: Chrysomelidae) dorsal habitus images at the genus level. In this study, a total of 888 habitus images belonging to 17 genera were obtained from a website on leaf beetles and five CNN models (ResNet-152, Alex-Net, DenseNet-201, VGG-16 and MobileNet-V2) were used to classify leaf beetle genera. Also, the classification performance of the models was compared. The most successful model was ResNet-152 with an accuracy rate of 97.74%. These results showed that Resnet-152 can be used to identify European leaf beetle genera. As a result of this study, it was concluded that as the number of images increases, the identification of leaf beetles at the genus level can be made more easily by using CNNs.
Alan : Fen Bilimleri ve Matematik
Dergi Türü : Uluslararası
Benzer Makaleler | Yazar | # |
---|
Makale | Yazar | # |
---|