Systematic Analysis of CNN and Its Optimization Algorithms Based on Object Detection

Abstract

Object detection is a very important research field in computer vision. In recent years, because of the development of Convolutional Neural Networks (CNNs), object detection methods are also more and more advanced. From the earliest Region-based Convolutional Neural Network (R-CNN), researchers proposed a two-stage detection idea, to You Only Look Once (YOLO) single-stage detection, which greatly improved the speed of object detection. There are also Transformer models like the Detection Transformer (DETR), which no longer require the complex post-processing steps of the original model, making detection more straightforward. This paper mainly sorts out how these CNN-based detection methods are developed step by step, analyzes their advantages and disadvantages, such as how the algorithm model achieves a significant improvement in accuracy and efficiency through multi-scale feature fusion, the challenges in detecting small objects and realizing real-time detection, and looks forward to possible future research directions, providing important theoretical and practical guidance for further improving model performance and expanding real-world applications.