Accurate New Zealand Wildlife Image Classification-Deep Learning Approach

Image classification is a major machine learning problem that has a wide range of applications in the real world. The Wellington Wildlife Camera Trap dataset contains images taken from vibration triggered cameras in sequences of three. State-of-the-art deep convolutional neural network (CNN) models, such as DenseNet-121 and ResNet-50, are unable to achieve the required accuracy of classification on this dataset. This research aims to improve the performance in multi-class classification tasks on the Wellington Dataset through a newly developed dual-input channel neural network. Our experiment results provide clear evidence that the new CNN model can achieve high accuracy and confidence on this challenging and scientifically important dataset. It is able to significantly reduce the amount of time required to manually classify wildlife images for conservation research in New Zealand.