Portal de Periódicos da CAPES

Bird Species Identification Based on Images Using Residual Network

2023; Springer International Publishing; Linguagem: Inglês

10.1007/978-981-19-9888-1_45

2367-3370

Haritha Potluri, Amuktha Vinnakota, Naga Pavan Prativada, Kalyan Chakravarti Yelavarti,

Wildlife Ecology and Conservation

Nowadays many of the bird species are found rarely, and it will be hard to classify the species of birds if found. For instance, in various cases, the birds may be of with various sizes, colors, forms and from the human viewpoints with various angles. This paper presents Bird species identification based on images using Residual Network (ResNet). In addition leverage pre-trained ResNet model is utilized as the pre-trained CNN networks with the base model for encoding the images. The process of determining the species of birds will involve several phases. The first stage involves an ideal dataset construction that is incorporated the images of various bird species. Next stage is image normalization where size of pixel will be processed. In this paper, the Caltech-UCSD Birds 200 [CUB-200-2011] data collection is utilized to train and test the presented model. 500 labeled data will be utilized for training purpose and 200 unlabeled data will be used to test the model. Final results show that ResNet model predicted at 96.5% of Accuracy and loss function with 15.06% of bird species. This approach will be performed over Linux operating system with the library of Tensor flow and utilizing the NVIDIA Geforce GTX 680 along with2 GB RAM.