研究目的
To propose a novel framework for aircraft detection in high resolution apron area in Synthetic Aperture Radar (SAR) images that combines the strength of location regression based convolutional neural network (CNN) framework and the salient features of target in SAR images.
研究成果
The proposed framework improves accuracy and detection efficiency a lot by taking full advantage of location regression based Fast-RCNN framework and the salient features of target in SAR images, and by exploring several strategies for SAR data augmentation to eliminate overfitting.
研究不足
The actual available SAR data is hardly enough to train a deep CNN properly, and subtle pose changes in the target may result in dramatic changes in imaging results.
1：Experimental Design and Method Selection:
The framework combines a Constant False Alarm Rate (CFAR) based target pre-locating algorithm with a location regression based Fast-RCNN for aircraft detection.
2：Sample Selection and Data Sources:
The data set acquired by the TerraSAR-X satellite in a resolution of
3：0 meters, containing 120 images with 2632 aircrafts. List of Experimental Equipment and Materials:
TerraSAR-X satellite data.
4：Experimental Procedures and Operational Workflow:
Data augmentation (translation, adding noise, rotation), target pre-locating using CFAR algorithm, model training (pre-training and fine-tuning phases), and target detection.
5：Data Analysis Methods:
Comparison of recall rate and average time cost (ATC) for different locating methods, and true positive rate (TPR) and false positive rate (FPR) for detection performance.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
