To solve the issues of low accuracy in detecting small targets in drone aerial images and the excessive computational complexity of the detection model, this paper proposes a drone small target detection method based on an enhanced version of YOLOv5. Firstly, in an endeavor to ameliorate the issue of undetected instances of small-scale targets, this study incorporates the Convolutional Block Attention Module (CBAM) into the feature extraction framework of YOLOv5s. This augmentation enhances the capacity for extracting features from diminutive targets, leading to increased accuracy in small target detection. Secondly, in response to the issue of high computational complexity in the model, this paper has improved the CBL module by replacing regular convolutions with Coord convolutions. This modification helps reduce the model's parameters and computational complexity, resulting in a more efficient and streamlined detection process. Lastly, this paper utilizes the EIOU Loss function to further improve the accuracy and convergence rate of target detection. The amalgamation of these improvements markedly enhances the model's performance. This paper conducted tests using the DOTA dataset, and the experimental results demonstrate the effectiveness of the CCE-YOLOv5s model. It outperforms the original model with a mAp (mean Average Precision) of 80%, represents a noteworthy 1% enhancement over the original algorithm. Moreover, the training time is reduced to 0.66 hours, a remarkable 43% reduction in comparison to the original algorithm. Overall, this improved model effectively improves detection accuracy while ensuring real-time performance, and significantly improves the training speed of the model.
CCE-YOLOv5s: An Improved YOLOv5 Model for UAV Small Target Detection
11.10.2023
3271558 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
UAV target detection algorithm based on improved YOLOv5s
SPIE | 2023
|An Improved YOLOv5-Based Small Target Detection Method for UAV Aerial Image
Springer Verlag | 2023
|