A Comparative Study of Different Optimisers on Convolutional Neural Networks for Bird Classification

The advantage of CNNs in classifying bird species through images has received public attention in the past several years due to the capability to get minute details from images. Still, CNN performance is greatly dependent on the type of optimiser selected during the training phase. This paper is focused on an experimental comparison of several optimisers like SGD, Adam and Rmsprop, as well as Adagrad, to see how they affect the classification accuracy and detection of bird species using CNNs. A large dataset of images featuring different bird species, which encompasses the different types of birds, is used to create robust results. Through extensive tests and evaluations, we estimate the speed of convergence, determination of stability, and classification performance of each optimiser and, finally, will assess the effectiveness of the algorithms. The findings of this work give great information about deciding on the best optimiser to fit the purpose of training a CNN, which the automation of a classification system depends on; the progress of such systems is one step closer to the automation of the bird species identification system.