Proposal of a Model for the Detection of Violence Against Animals Through Convolutional Neural Networks

Animal violence is a global problem that affects millions of animals every year. It can manifest itself in various forms, such as abandonment, physical abuse, and commercial exploitation. Animal violence has a negative impact on animal welfare, but it can also have consequences for society at large. Detecting animal violence is a major challenge. Traditional methods, such as bystander reporting, may be insufficient to identify all cases of abuse. Automated detection methods, such as machine vision, may be a valuable tool to improve the detection of animal violence. This study evaluated the effectiveness of convolutional neural networks, coupled with pose estimation, in detecting animal violence in videos. A dataset of animal violence videos was created and preprocessed to improve model performance. Several detection models, such as YOLOV7, VGG-16, VGG-19, GoogleNet and ResNet, were implemented, of which, they were evaluated based on their accuracy, precision, and sensitivity. Accordingly, the results showed that YOLOV7, in combination with pose estimation, obtained the best performance metrics. The model demonstrated an accuracy of 93.49%, precision of 98% and sensitivity of 88%. suggest that YOLOV7, together with pose estimation, is an effective method for detecting animal violence. The model provides a comprehensive approach to identify violent behaviors towards animals and can contribute to the automation and improvement of the identification of such behaviors.