A Comparative Study of Optimizers on Non-pretrained CNN Models for Stray Animal Surveillance System

Abstract

The stray animals sighted on the Indian vehicular roads cause traffic congestions and lead to major road accidents. Due to the menace, many people and animals get serious injuries and even lose their life. The attacks on the humans, damaging of properties and spreading of dangerous diseases such as rabies are the other major concerns due to the increase in stray animals. The present study performs detection and classification of the stray animals on the vehicular streets. The dataset comprises of 500 images of vehicular roads with and without stray animals. The classification accuracy of the surveillance system is compared between Stochastic Gradient Descent (SGD) and Adam optimizer for a 3-layer and 4-layer CNN having different batches. The study aims to act as a surveillance system on the roads for detecting the presence of stray animals. Timely detection and relocation of stray animals can help in preventing the fatalities and spread of diseases. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.