We present and discuss results of the application of a deep convolutional network model developed for the automatic recognition of images of insects. The network was trained using transfer learning on an architecture called MobileNet, specifically developed for mobile applications. To fine tune the model, a grid-search on hyperparameters space was carried out reaching a final accuracy of 98.39% on 11 classes. Fine-tuned models were validated using 10-fold cross validation and the best model was integrated into an Android application for practical use. We propose solving the “open set” problem through feed-back collected with the application itself. This work also led to the creation of a well-structured image dataset of some important species/genera of insects.
Insects Image Classification Through Deep Convolutional Neural Networks
Teresa Bonacci;
2021-01-01
Abstract
We present and discuss results of the application of a deep convolutional network model developed for the automatic recognition of images of insects. The network was trained using transfer learning on an architecture called MobileNet, specifically developed for mobile applications. To fine tune the model, a grid-search on hyperparameters space was carried out reaching a final accuracy of 98.39% on 11 classes. Fine-tuned models were validated using 10-fold cross validation and the best model was integrated into an Android application for practical use. We propose solving the “open set” problem through feed-back collected with the application itself. This work also led to the creation of a well-structured image dataset of some important species/genera of insects.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
