研究目的
To improve the accuracy and speed of object detection in human body THz images by combining preprocessing and structure optimization of Faster R-CNN.
研究成果
The R-PCNN method can improve the accuracy and speed of object detection in THz human body security images, achieving a detection accuracy of 84.5% in dense flow scenes with an average detection time of less than 20 milliseconds for each image. Future work includes reconstructing the super-resolution of the terahertz image and enriching the THz image data sets.
研究不足
The low resolution and signal-to-noise ratio of terahertz images and the lack of diversity in terahertz image data sets are the main limitations.
1：Experimental Design and Method Selection:
The study proposes a Faster R-CNN detection framework (R-PCNN) that combines preprocessing and structural optimization for object detection in THz human body security images.
2：Sample Selection and Data Sources:
The objects are divided into two categories: cutting tools and mobile phones, with a total of 39 knives, 41 mobile phones, and 38 terahertz images of human body security.
3：List of Experimental Equipment and Materials:
Hardware configuration includes an i7-6700K *8 CPU,
4：0GHz main frequency, 16GB memory, and a GTX750Ti/GTX1080 GPU. Software configuration was Open CV-11 + Python 7 + Caffe +Faster R-CNN. Experimental Procedures and Operational Workflow:
The original image is preprocessed for denoising and enhancement, then detected by the trained network.
5：Data Analysis Methods:
The performance results regarding the detection accuracy and speed are compared between various algorithms.
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
