Search

The terahertz (THz) wave can penetrate many nonmetallic materials, such as paper, plastic, and clothing, making it highly valuable in the field of security checks. However, the low contrast between target objects and the background in THz security inspection images poses challenges for object semantic segmentation tasks. Although various solutions are proposed to address these issues, achieving a balance among low localization and segmentation performance, training data, and model complexity remains a problem. To tackle this problem, we propose an instance segmentation method called YOLO-based instance segmentation method (IYOLO) to locate and segment concealed hazardous objects in THz security inspection images. A channel-spatialized Transformer (CST) module is developed to strengthen backbone feature extraction capabilities by combining global feature learning with optimized channel and spatial features. Moreover, an enhanced upsample block (EUB) is designed to integrate channel operations, encoder-decoder structure, and feature enhancement to improve feature maps. In addition, a shuffle convention (Conv) layer (SCLayer) is proposed in the segmentation head to optimize model parameters and complexity while enhancing the perception and accuracy of the model. Many experiments are conducted on public datasets. The results demonstrate that IYOLO outperforms other state-of-the-art (SOTA) methods in the concealed hazardous object instance segmentation task of THz security inspection images. Especially, in complex scenes, IYOLO shows excellent segmentation ability and effectively avoids false positives and false negatives. Ablation experiments confirm the effectiveness of the proposed module, which improves detection performance while maintaining moderate model complexity. Moreover, compared with other methods, the IYOLO model exhibits superior size and computational complexity, making it more practical for real-world security inspection applications. The code of the proposed method is available at https://github.com/xiaokai11/IYOLO.

Cylindrical shaft parts are critical in optical applications where performance is highly dependent on surface quality and therefore require highly refined surface characteristics. When using a multibeam angle sensor (MBAS) for cylindrical shaft parts inspection, the inspection accuracy is reduced due to subspot distortion caused by long camera exposure times and natural light interference. On this basis, a multispot image denoising and correction network (MSIDNet) is proposed, aiming at processing multispot images by denoising and correction techniques, in addition to combining Gaussian fitting methods (GFMs) to improve subspot localization accuracy to the subpixel level. A subspot offset determination method based on moving window matching subspots is proposed, which provides a reliable basis for the 3-D reconstruction of cylindrical shaft parts. Experimental results show that MSIDNet can effectively remove noise interference and achieve accurate extraction of subpixel level subspot centers. Cylindrical shaft parts with radii of 10, 20, and 22 mm were measured using this method, and the results showed that their standard deviations (stds) were 0.421, 0.207, and $0.071~\mu $ m, respectively.

Defect detection in multilayer ceramic capacitors (MLCCs) is critical for ensuring the production quality. However, surface defect detection in MLCC faces challenges, such as multiscale defects, indistinct boundaries, and limited availability of defect samples. To address these issues, this article proposes an automated defect detection system based on machine vision. The system includes a microscopic imaging unit (MIU) for defect image capture, a motion control unit (MCU) for MLCC movement control, and an image processing unit (IPU) for efficient MLCC surface defect segmentation. In addition, a lightweight multiscale network (LMS-net) is introduced for defect segmentation. LMS-net aims to effectively capture image context information, extract richer semantic details, and handle class imbalance. It incorporates a deep feature extraction module (DFEM) to improve the feature extraction capabilities and reduce computational burden, a dual attention (DA) to enhance attention to smaller defect areas, and a cascaded pyramid feature fusion module (CPFFM) to incorporate context information at different scales. In addition, a combined loss function is used to handle the class imbalance during the training process. Experimental results demonstrate the outstanding performance of the proposed network on the MLCC surface defect dataset, achieving a 94.14% dice similarity coefficient (DSC), a 67.01% mean intersection over union (MIoU), and 86.41% mean pixel accuracy (MPA), all with a smaller parameter.

The appearance of defects in a multilayer ceramic capacitor (MLCC) adversely affects its performance and reliability. Thus, detecting these defects during MLCC production is imperative. However, this task faces numerous challenges, such as significant variations in the shape and size of defects, indistinct defect boundaries, and the inefficiency of manual detection methods. To address these issues, this article proposes the RSE-YOLO model for identifying, localizing, and classifying defects in MLCC images. We design a novel backbone structure, namely the residual coordinate weighted convolutional network (RCWCNet), which possesses enhanced feature information extraction capabilities for accurately locating defect areas. Additionally, we introduce the space attention pyramid pooling module (SAPPM) to achieve a weighted fusion of local and global feature information. Furthermore, the ECA-PAN is employed as the model’s neck structure to facilitate the fusion of feature information at different scales, improving the model’s generalization ability in multiscale defect detection. Experimental results demonstrate that the RSE-YOLO model exhibits excellent performance on the MLCC dataset, achieving an mAP50 of 93.9%, mAP50-95 of 63.2%, F1 of 90.9%, and a frame rate of 57 FPS, meeting the requirements for the task of MLCC appearance defect detection.

Infrared thermal imaging (ITI) technology that has unique advantages in obtaining target information at night and in harsh weather conditions has been widely used in the military, security, environmental monitoring, and other fields. Existing visible light object detection methods often face challenges of high false alarm rates and low accuracy when detecting small infrared objects. Moreover, current small infrared object detection methods mostly require much detailed annotation data, and it is challenging to visually depict the positions of small targets on the original complex background images. To address these issues, we propose a multibranch backbone and adaptive spatial feature fusion detection head-based you only look once (YOLO) network called MBFormer-YOLO to achieve the high-accuracy detection of small infrared targets. We design a multibranch backbone with structural re-parameterization capability called MBFormer, which has strong capabilities to fit object features. In the neck part, MBFormer-YOLO utilizes convolutional block attention modules (CBAMs) and C3-SwinTransformer (C3-SwinT) modules to suppress noise and redundant information. We develop a four-level adaptive spatial feature fusion detection head (ASFFDH) to improve the detection accuracy of the proposed network. Many experiments are conducted on the SIRST-V2 dataset to validate the effectiveness of MBFormer-YOLO. The results show that MBFormer-YOLO achieves an 11.3% increase in AP50 and 2% in AP50-95 over the baseline model YOLOv8n and surpasses other advanced object detection models.