IMAGE BASED BIRD SPECIES IDENTIFICATION USING DCNN ARCHITECTURE

In our everyday lives, there are occasional encounters with various species of birds, and if a grouping of bird species is established, it can sometimes involve a challenging process. Typically, diverse types of birds can be identified based on their vibrant colors, unique physical structures, and different angles from an observer’s perspective. While photographs offer distinct and robust dissimilarities that aid in recognizing bird breeds more conveniently than audio recordings, the innate ability to appreciate birds through captured images is even more advantageous. To facilitate this purpose, a cutting-edge dataset called the Caltech-UCSD Birds 200 (CUB-2000-2011) dataset is employed for both training and testing in a device known as a gadget. This device utilizes the powerful DCNN (Deep Convolution Neural Network) algorithm, which excels at modeling graph-structured data. By converting an image into a slate-scale format and extracting signatures through tensor inflow, the device creates distinct comparison nodes using image processing techniques. By establishing a database of standardized image features for bird species and employing a correspondence comparison algorithm, this project has successfully achieved favorable outcomes in practical applications.