Development of a Deep learning Image Classification Network based on Vehicular Collision using Instance Mask Guided Attention.

This paper presents Understanding vehicular collisions from visual data is essential for improving the safety and decision-making capability of autonomous driving systems.This work introduces a deep learning-based image classification framework that incorporates an Instance Mask-Guided Attention (IMGA) mechanism to enhance the network's focus on collision-related regions.In the proposed approach, attention learning is explicitly supervised using instance masks generated for vehicles involved in collisions, enabling the model to concentrate on the most discriminative spatial features while minimizing background interference.A dedicated vehicular collision image dataset was developed to support training and evaluation.Furthermore, a modified VGG-19 architecture integrated with IMGA was designed, achieving an accuracy of 97.94% on the constructed dataset and surpassing several contemporary deep learning classifiers.Experimental results confirm that even a relatively lightweight backbone can deliver high-performance classification when guided by instance masks, eliminating the need for deeper and computationally expensive architectures.