A Hybrid Approach for Target Discrimination in Remote Sensing: Combining YOLO and CNN-Based Classifiers

Abstract

With the increase in image production in recent years, there has been significant progress in the application of deep learning algorithms across various domains. Convolutional neural networks (CNNs) have been increasingly employed in remote sensing, covering all stages of target discrimination according to Johnson’s criteria (detection, recognition, and identification). These CNNs are applied in many conditions and imagery from many types of sensors. In this study, we explored the use of the YOLO-v8 method, the latest version of the You Only Look Once (YOLO) family of object detection models, in conjunction with CNN architectures and supervised learning algorithms. This approach was applied to detect, recognize, and identify targets in videos captured by optical sensors, considering varying resolutions and conditions. Additionally, our research investigated the use of two CNN architectures, Inception-v3 and VGG-16, to extract relevant information from the images. The attributes obtained from the CNNs were used as input for three classification algorithms: multilayer perceptron (MLP), logistic regression, and support vector machine (SVM), thereby completing the target discrimination process. It is worth noting that in the combination of Inception-v3 and MLP, we achieved an average accuracy of 90.67%, thus completing the target discrimination process.