A New Efficient Classifier for Bird Classification Based on Transfer Learning

Abstract

Te problem of classifcation of diferent species of birds in the images is relevant in the modern world. Te essence is to automatically determine the species of birds depicted in the photo using artifcial intelligence. It is important for several areas of human life, in particular for nature conservation, environmental research, education, and ecotourism. Te development of a classifer, based on deep learning methods, can help to efectively solve the problem of classifying diferent species of birds, ensuring high accuracy. Te work developed a new efective algorithm for the classifcation of diferent species of birds in the images. An optimal model architecture was built using the transfer learning approach. A dataset of 525 bird species was analyzed and preprocessed in detail. Te model training process was carried out, which includes two phases: only the upper layers are deactivated and the last 92 layers of the pretrained EfcientNetB5 model (not including BatchNormalization layers) and the top layers are activated. As a result, efcacy indicators were obtained: accuracy = 98.86%, precision = 0.99, recall = 0.99, and F1 score = 0.99, which showed improvement in comparison with modern research. Te developed classifer is best suited for such areas of human life as education and ecotourism because for them the number of diferent species of birds that the classifer can determine is very important.