Bird Image Dataset Classification using Deep Convolutional Neural Network Algorithm

The conventional system that is now in place was majorly used in practical problems, especially in classification problems such as lack of adaptability, weak classification accuracy, and unsatisfying results. Deep Learning (DL) was offered as a solution that could address such problems and increase accuracy as in image classification. This research explains a new intelligent system for bird type determination that estimated from bird shape images based on external appearance using multitasking and Convolutional Neural Network (CNN) is proposed. The importance of the proposed model that can obtain a good and satisfying accuracy from using CNN which performed well for the most abundant class through extracting important features. The typical structure of the proposed system consists of two stages: The first stage is the pre-processing stage and the classification stage used to give, which are gives the type and the name of the bird image. The proposed system used (BIRDS 400) species image classification. The train directory contains 58388 training images and the test directory contains 2000 test images both of size 224 × 224 × 3 in jpg format. The directories in train and test images are partitioned into 400 sub directories, one for each species (bird name). The Proposed Convolution Neural Network (CNN), which has 15 layers and gives high percentages of accuracy Precision near to 0.99%, Recall scale is 0.95% and F-measure 93%, the proposed deep CNN algorithm has proven its efficiency in classifying the bird dataset by comparing it with other machine learning classification algorithms which are ((NaiveBase, Random Forest, Decision Tree, and K-NN Algorithm).