Evaluation of Pre-Trained Convolutional Neural Networks for the Development of Web/mobile-Based Bird Species Identification Applications under Indian Bioresource Information Network (IBIN) Framework

An early indication of changes in ecosystems can be noticed by the presence and behaviour of a few species that are highly sensitive to the change in the environment. Bird species are excellent indicators of changing ecosystems because they react fast to changes in their surroundings. Therefore, knowledge of all bird species and identifying their types are required to monitor the changing habitats and ecosystems. The performances of various pre-trained convolution neural networks (CNN) namely, MobileNet, MobileNetV2, NASNetMobile, and EfficientNets (version B0, B1, B2, B3, B4), in identifying Indian bird species using visual scene classification are evaluated in this study. As part of the Indian Bioresource Information Network (IBIN), this evaluation will aid in the development of Web/Mobile-Based Bird Species Identification Applications. The results indicate that EfficientNet B4 outperformed other architectures, however, the computational complexity is higher among all the selected CNN models. The EfficientNetB0 has demonstrated comparable performance with computational complexity comparable to MobileNet and NASNet and can be used in mobile platforms.