Detection of Objects in Remotely Sensed Images using an Explainable Convolutional Neural Network Enhanced with Tetralet Attention Optimized by Canis Hunt

Object or Target identification has perennially represented a focal point of investigation within the realm of computer vision; particularly the task of detection in satellite imagery has emerged as a persistently intricate research domain. Despite various traditional studies attempting to address this issue, they have encountered limitations when handling high-resolution satellite images. The present study introduces a novel approach to object detection, referred to as the CHunt-TetraExNN. The primary objective of this proposed model is to achieve precise object detection utilizing airborne image inputs. The Tetralet attention mechanism integrated into the system consists of a triad of attention components paired with positional attention, facilitating the precise delineation of attention-based attributes crucial for object detection tasks Furthermore, the investigative framework utilizes the Canis Hunt Optimization method, combining the adaptable characteristics and predatory behaviors of Lapins and Latrans species from the Canis Family. As a result, the framework provides precise estimation results in the field of object identification in airborne images, demonstrating an Accuracy rate of 96.48%, Precision of 94.60%, Recall of 96.52%, and an F1 score of 95.60%.