Anomaly detection has various applications in the industrial field. However, most current studies in this field are idealized such that they can only address anomaly detection in simple scenarios. For instance, most studies assumed that all detection objects are centered and typically have a homogeneous monochromatic background, with little spatial position change across different images. Such an assumption makes it difficult to apply the research findings to complex industrial production scenarios. To address the limitations of existing idealized approaches, we propose a frequency-guided image reconstruction method with multi-level feature combination (MLFC) and multi-scale feature fusion (MSFF) for anomaly detection. The overview of the proposed method is as follows: First, we employ a mixed anomaly generation strategy to generate anomalous samples for training. Second, before feeding the images into the reconstruction network, we introduce a frequency selection module. This module assigns different frequency weights to different frequency components, to suppress the tendency toward identity mapping of the anomalies of the network. Finally, we design a reconstruction network based on U-Net by employing three MLFC modules and one MSFF module. The MLFC modules in the skip-connections adaptively fuse features from multiple encoder levels using channel attention. The MSFF module thereafter fuses the multi-scale features from the decoder to enhance the representational capability of the decoder output features. The experimental results demonstrate that even with a simple U-Net as the reconstruction network, the proposed method achieves state-of-the-art performance on the metal parts defect detection dataset. Specifically, it reaches an image-level area under the receiver operating curve (AUROC) of 94.91% and a pixel-level AUROC of 97.88%, setting a new benchmark.
Footprint is a human biological trait with a high extraction rate from crime scenes. It plays an essential role in criminal investigation and security operations. The majority of footprint studies are currently conducted under controlled conditions. However, footprints formed under natural walking are affected by physiological and behavioral characteristics such as posture and walking state, and have variability, which makes it difficult to accurately identify footprints. In this paper, multi-state barefoot pressure images under natural walking state are taken as the research object, and the selective attention network is used for high precision recognition. It can provide a theoretical foundation as well as technical support for later comparison and identification of footprints on-site.
To evaluate the rehabilitation effect of elderly people during walking training on treadmills, a detection system capable of acquiring gait features was designed. Install the flexible force-sensitive sensor on the treadmill table, collect sensor data through STM32F205 microcontroller, and use W5300 Ethernet chip and STM32F205 microcontroller to build a TCP/IP network communication platform to ensure that sensor data can be synchronously transmitted to the host computer through the network port in real-time. The gait characteristics of the elderly are obtained by calculating the sensor data obtained by the host computer. System tests show that the gait characteristics obtained when walking on a treadmill are consistent with previous research results, thus verifying the accuracy of the detection system.
Ski jumping is a fast and wide-range motion. Although wearable devices are available to analyze the motion, it is cumbersome and difficult to implement. Since video data is relatively simple to obtain, this paper proposes a video-based method for estimating the pose of ski jumpers. In this method, we use a high-speed camera as a video data collector, and use Simi Motion software to convert the video into frames and manually annotate keypoints. The video data of three athletes is used to build the training set, and another is used to build the test set. In addition, we use High-Resolution Net (HRNet) to transfer the learning of feature knowledge from the public dataset COCO2017 to the task of ski jumpers pose estimation. The experiments show that under the real labeled bounding box, an average precision of 84.6% is obtained, which is higher than other mainstream human pose estimate methods.
Footprints are important information at the crime scene and play an important role in the field of criminal investigation. At present, the research on footprints mainly focuses on barefoot footprints, but the main thing obtained at the crime scene is shoeprints. How to mine barefoot footprints through shoeprints is one point of the key problems in the field of footprint recognition. This paper takes optical footprints as the research object, collects 95 people’s cloth shoeprints and barefoot footprints, and proposes a generative adversarial network which combines self-attention modules and multiscale discriminator (SM-GAN). The self-attention module is added to the generator, which enables the network to focus on the association between footprint structures. The discriminator uses a multiscale discriminator structure, which improves the generation effect of the generated image in the global and local areas. The experimental results show that the method proposed in this paper has a better effect of generating footprints than traditional image-to-image translation methods.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.