Drones have become increasingly prevalent across various fields, including environmental monitoring and emergency rescue, due to their flexibility in deployment and ease of exploration. However, in practical applications, traditional methods that rely on ground operators to manually analyze and identify targets from returned images often fail to provide accurate and real-time responses. To address this challenge, integrating deep learning techniques for real-time target detection and recognition has emerged as a critical focus for small unmanned aerial vehicles (UAVs). This study introduces an object detection approach that integrates real drone platforms and employs the YOLOv5 object detection network to identify objects in RTSP video streams transmitted by drones. The trained YOLOv5 model is deployed on a laptop, where it utilizes FFmpeg to receive real-time video stream data and perform object detection and classification. The proposed method has yielded promising results in demonstration experiments, validating its effectiveness and reliability in providing real-time target identification based on drone video feedback.