E-CNN: Ensemble-Based Convolutional Neural Network for Animal Species Identification

As emerging of technology and increasing population requires the urbanization in the most of the countries. The urbanization leads to destroy the large trees, plants and bushes that provide the shelter, food and living ecosystem. Due to this urbanization changes, various animal species are in need of searching the new shelter. Hence, the forest management should identify the various animal species using unmanned aerial devices, and transfer them into right living shelter. This study proposed the Ensemble Convolutional Neural Network (E-CNN) for detecting and classifying various animal species. The modified light weight MobileNetV2 proposed to classify the various animal species. Four CNN models InceptionResNetV2, DenseNet121, and MobileNetV2 outputs are combined and applied the ensemble method to improve the better performance. The transfer learning is applied in top layers and few convolution layers to improve the training time and accuracy. The proposed model achieved the accuracy of 98.2 % in classifying the various animal species. This research considers 6 animal species from the entire dataset that contains 90 species. The ablation study proved that the ensemble learning performs better comparing to the individual CNN model.