| 摘要: |
| 现有飞行训练科目识别方法在复杂机动动作识别中存在滞后性，忽视了现代数字技术的应用与最新规章制度的更新，其核心挑战在于如何降低匹配误差率并提升运算速度。针对上述问题，设计并构建了一种混合自动识别框架（SURF-FLANN-RANSAC Hybrid Algorithmic Framework,SFR）。首先，采用改进的加速稳健特征（Speeded Up Robust Features，SURF）算法实现驾驶舱场景内的图像特征提取与匹配。其次，引入近似最近邻快速搜索库（Approximate Nearest Neighbors，FLANN）匹配器以加速特征匹配过程，提升特征匹配效率与精度。然后，基于随机采样一致性（Random Sample Consensus，RANSAC）算法消除误匹配问题，增强算法整体鲁棒性。在大坡度盘旋、懒“8”、急上升转弯3个典型飞行训练场景自建数据集上的实验结果表明，该算法的识别精度分别为94.58%、62.95%、86.72%，与表现次佳的算法相比，分别提升了1%、20%、4%，且处理速度上实现了显著改善，为飞行训练的智能化管理提供了强有力的技术支持。 |
| 关键词: 飞行训练动作识别 图像特征提取 特征点匹配 加速稳健特征（SURF） |
| DOI：10.20079/j.issn.1001-893x. 250408001 |
|
| 基金项目:中央高校基本科研业务费专项资金资助（24CAFUC03039）;西藏自治区科技重大专项资金资助（XZ202101ZY0017G） |
|
| Automatic Recognition of Flight Training Scenarios Based on an Optimized SURF Algorithm |
| ZHANG Yalan,LIU Fang,LIU Weijie,WEI Yongchao |
| (School of Computer Science,Civil Aviation Flight University of China,Guanghan 618307,China) |
| Abstract: |
| The existing flight training subjects identification methods exhibit latency in recognizing complex maneuvering actions,often overlooking the potential of modern digital technologies and the latest regulatory updates.The core challenge lies on reducing matching error rates while enhancing computational efficiency.To address these issues,a SURF-FLANN-RANSAC hybrid algorithmic framework(SFR) is proposed.First,an improved Speeded Up Robust Features(SURF) algorithm is utilized to extract and match image features within cockpit environments.To further enhance the efficiency and accuracy of feature matching,the Fast Library for Approximate Nearest Neighbors(FLANN) matcher is integrated into the process.Additionally,the Random Sample Consensus(RANSAC) algorithm is applied to eliminate mismatches,thereby improving the overall robustness of the system.Experimental results on a self-built dataset comprising three typical flight training scenarios,steeply banked turns,lazy eights,and chandelles,demonstrate that the proposed algorithm achieves recognition accuracies of 94.58%,62.95%,and 86.72%,respectively.Compared with that of the second-best performing algorithm,these results represent improvements of 1%,20%,and 4%,respectively,along with a significant enhancement in processing speed,offering strong technical support for the intelligent management of flight training. |
| Key words: identification of flight training action image feature extraction feature point matching speeded up robust features(SURF) |
