Search Results (594)

The mid-wave infrared (MWIR) spectral range can provide a larger bandwidth for optical sensing and communication when the near-infrared band becomes congested. This range of thermal signatures can provide more information for digital imaging and object recognition, which can be unraveled from polarization-sensitive detection by integrating the metasurface of the subwavelength-scale structured interface to control light–matter interactions. To enforce the metasurface-enabled simultaneous detection and parallel analysis of polarization states in a compact footprint for 4-micron wavelength, we designed a high-contrast germanium metasurface with an axially asymmetric triangular nanoantenna with a height 0.525 times the working wavelength. First, we optimized linear polarization separation of a 52-degree angle with about 50% transmission efficiency, holding the meta-element aspect ratio within the 3.5–1.67 range. The transmission modulation in terms of periodicity and lattice resonance for the phase-gradient high-contrast dielectric metasurface in correlation with the scattering cross-section for both 1D and 2D cases has been discussed for reducing the aspect ratio to overcome the nanofabrication challenge. Furthermore, by employing the geometric phase, we achieved 40% and 60% transmission contrasts for the linear and circular polarization states, respectively, and reconstructed the Stokes vectors and output polarization states. Without any spatial multiplexing, this single metasurface unit cell can perform well in the division of focal plane Stokes thermal imaging, with an almost 10-degree field of view, and it has an excellent refractive index and height tolerance for nanofabrication. Full article

Addressing the problem of inadequate environmental detection in the process of optimizing search for unmanned surface vehicles (USVs) by a heuristic algorithm, this paper proposes a comprehensive visual perception method that combines a lightweight convolutional neural network (CNN) with the USV’s real-time heading angle. This method employs a multi-feature input CNN with residual learning blocks, which takes both the current local environmental images and heading angle features as inputs to identify the complexity of the local environment with higher accuracy and a smaller load size. Meanwhile, human expertise is incorporated to classify labels through a majority voting system, thereby making the model’s perceptual classification more intuitive and allowing it to possess a human-like comprehensive perception ability compared to systems with classification methods with several parameters. Subsequently, this identification result can be used as feedback for the heuristic algorithm to optimize and plan the USV’s path. The simulation results indicate that the developed model achieves an 80% reduction in model size while maintaining an accuracy exceeding 90%. The proposed method significantly improves the environment recognition capability of the heuristic algorithm, enhances optimization search efficiency, and increases the overall performance of path planning by approximately 21%. Full article

The structural parameters of the liquid sheet represent a significant factor influencing the atomization performance, and its measurement is an important part of the agrochemical atomization study. Currently, the measurement predominantly relies on commercial software with manual operation, which is labor intensive and inefficient. In this study, deep learning methods with high-speed photographing were employed to measure the structural parameters of the liquid sheet of hydraulic nozzles with different atomization modes. The LM-YOLO liquid sheet structure recognition model was constructed to recognize the liquid sheet and perforations. Based on the recognition results, a method is designed to calculate several key parameters, including the breakup length, the liquid sheet area, the spray angle, the average number of perforations, and the average perforation area. A comparative scrutiny of the assorted liquid sheet structural parameters under different experimental conditions was also implemented. Based on the constructed model, a recognition accuracy of 81.0% for the liquid sheet structure of the LU nozzle (a classical hydraulic nozzle with high liquid sheet integrity) and 71.3% for the IDK nozzle (an air-induced hydraulic nozzle with a certain amount of bubbles in the liquid sheet) was achieved. The liquid sheet structure was measured based on the recognition results. It was found that the pressure has a significant impact on the structural parameters of the liquid film. For the LU120-03 nozzle, the breakup length of the liquid film decreases from 48.96 mm to 39.05 mm as the pressure increases. In contrast, for the IDK120-03 nozzle, the breakup length exhibits fluctuating changes, with a peak value of 29.65 mm occurring at 250 kPa. After adding silicone adjuvant, the breakup length and area of the liquid film generally decrease. The variation trends of the measured structural parameters under different experimental conditions are consistent with the trends of the data in previous relevant research by other scholars. This study provides a new method for measuring out the structural parameters of the liquid sheet, and it has potential application in related fields. Full article

With the advancement of machine vision technology, deep learning and image recognition have become research hotspots in the non-destructive testing of agricultural products. Moreover, using machine vision technology to identify different ripeness stages of fruits is increasingly gaining widespread attention. During the ripening process, bananas undergo significant appearance and nutrient content changes, often leading to damage and food waste. Furthermore, the transportation and sale of bananas are subject to time-related factors that can cause spoilage, necessitating that staff accurately assess the ripeness of bananas to mitigate unwarranted economic losses for farmers and the market. Considering the complexity and diversity of testing environments, the detection model should account for factors such as strong and weak lighting, image symmetry (since there will be symmetrical banana images from different angles in real scenes to ensure model stability), and other factors, while also eliminating noise interference present in the image itself. To address these challenges, we propose methods to improve banana ripeness detection accuracy under complex environmental conditions. Experimental results demonstrate that the improved ESD-YOLOv9 model achieves high accuracy in these conditions. Full article

Childhood glaucoma encompasses a group of rare but severe ocular disorders characterized by increased intraocular pressure (IOP), posing significant risks to vision and quality of life. Primary congenital glaucoma has a prevalence of one in 10,000–68,000 people in Western countries. More worryingly, it is responsible for 5–18% of all childhood blindness cases. According to the Childhood Glaucoma Research Network (CGRN), this spectrum of disease is classified into primary glaucoma (primary congenital glaucoma and juvenile open-angle glaucoma) and secondary glaucomas (associated with non-acquired ocular anomalies, non-acquired systemic disease, acquired conditions, and glaucoma after cataract surgery). They present very specific ocular characteristics, such as buphthalmos or progressive myopic shift, corneal modifications such as Haab striae, corneal edema or increased corneal diameter, and also glaucoma findings including high intraocular pressure, specific visual fields abnormalities, and optic nerve damage such as increased cup-disc ratio, cup-disc ratio asymmetry of at least 0.2 and focal rim thinning. Surgical intervention remains the cornerstone of treatment, and initial surgical options include angle surgeries such as goniotomy and trabeculotomy, aimed at improving aqueous outflow. For refractory cases, trabeculectomy and glaucoma drainage devices (GDDs) serve as second-line therapies. Advanced cases may require cyclodestructive procedures, including transscleral cyclophotocoagulation, reserved for eyes with limited visual potential. All in all, with appropriate management, the prognosis of PCG may be quite favorable: stationary disease has been reported in 90.3% of cases after one year, with a median visual acuity in the better eye of 20/30. Immediate recognition of the specific signs and symptoms by caregivers, primary care providers, and ophthalmologists, followed by prompt diagnosis, comprehensive surgical planning, and involving the caregivers in the follow-up schedule remain critical for optimizing outcomes in childhood glaucoma management. Full article

This study presents a method for the active control of a follow-up lower extremity exoskeleton rehabilitation robot (LEERR) based on human motion intention recognition. Initially, to effectively support body weight and compensate for the vertical movement of the human center of mass, a vision-driven follow-and-track control strategy is proposed. Subsequently, an algorithm for recognizing human motion intentions based on machine learning is proposed for human-robot collaboration tasks. A muscle–machine interface is constructed using a bi-directional long short-term memory (BiLSTM) network, which decodes multichannel surface electromyography (sEMG) signals into flexion and extension angles of the hip and knee joints in the sagittal plane. The hyperparameters of the BiLSTM network are optimized using the quantum-behaved particle swarm optimization (QPSO) algorithm, resulting in a QPSO-BiLSTM hybrid model that enables continuous real-time estimation of human motion intentions. Further, to address the uncertain nonlinear dynamics of the wearer-exoskeleton robot system, a dual radial basis function neural network adaptive sliding mode Controller (DRBFNNASMC) is designed to generate control torques, thereby enabling the precise tracking of motion trajectories generated by the muscle–machine interface. Experimental results indicate that the follow-up-assisted frame can accurately track human motion trajectories. The QPSO-BiLSTM network outperforms traditional BiLSTM and PSO-BiLSTM networks in predicting continuous lower limb motion, while the DRBFNNASMC controller demonstrates superior gait tracking performance compared to the fuzzy compensated adaptive sliding mode control (FCASMC) algorithm and the traditional proportional–integral–derivative (PID) control algorithm. Full article

Currently, various deep learning methods can assist in medical diagnosis. Coronary Digital Subtraction Angiography (DSA) is a medical imaging technology used in cardiac interventional procedures. By employing X-ray sensors to visualize the coronary arteries, it generates two-dimensional images from any angle. However, due to the complexity of the coronary structures, the 2D images may sometimes lack sufficient information, necessitating the construction of a 3D model. Camera-level 3D modeling can be realized based on deep learning. Nevertheless, the beating of the heart results in varying degrees of arterial vasoconstriction and vasodilation, leading to substantial discrepancies between DSA sequences, which introduce errors in 3D modeling of the coronary arteries, resulting in the inability of the 3D model to reflect the coronary arteries. We propose a coronary DSA video sequence keyframe recognition method, CDKD-w+, based on w+ space encoding. The method utilizes a pSp encoder to encode the coronary DSA images, converting them into latent codes in the w+ space. Differential analysis of inter-frame latent codes is employed for heartbeat keyframe localization, aiding in coronary 3D modeling. Experimental results on a self-constructed coronary DSA heartbeat keyframe recognition dataset demonstrate an accuracy of 97%, outperforming traditional metrics such as L1, SSIM, and PSNR. Full article

Food recognition from images is crucial for dietary management, enabling applications like automated meal tracking and personalized nutrition planning. However, challenges such as background noise disrupting intra-class consistency, inter-class distinction, and domain shifts due to variations in capture angles, lighting, and image resolution persist. This study proposes a multi-stage convolutional neural network-based framework incorporating a boundary-aware module (BAM) for boundary region perception, deformable ROI pooling (DRP) for spatial feature refinement, a transformer encoder for capturing global contextual relationships, and a NetRVLAD module for robust feature aggregation. The framework achieved state-of-the-art performance on three benchmark datasets, with Top-1 accuracies of 99.80% on the Food-5k dataset, 99.17% on the Food-101 dataset, and 85.87% on the Food-2k dataset, significantly outperforming existing methods. This framework holds promise as a foundational tool for intelligent dietary management, offering robust and accurate solutions for real-world applications. Full article

With the rapid development of AI algorithms and computational power, object recognition based on deep learning frameworks has become a major research direction in computer vision. UAVs equipped with object detection systems are increasingly used in fields like smart transportation, disaster warning, and emergency rescue. However, due to factors such as the environment, lighting, altitude, and angle, UAV images face challenges like small object sizes, high object density, and significant background interference, making object detection tasks difficult. To address these issues, we use YOLOv8s as the basic framework and introduce a multi-level feature fusion algorithm. Additionally, we design an attention mechanism that links distant pixels to improve small object feature extraction. To address missed detections and inaccurate localization, we replace the detection head with a dynamic head, allowing the model to route objects to the appropriate head for final output. We also introduce Slideloss to improve the model’s learning of difficult samples and ShapeIoU to better account for the shape and scale of bounding boxes. Experiments on datasets like VisDrone2019 show that our method improves accuracy by nearly 10% and recall by about 11% compared to the baseline. Additionally, on the AI-TODv1.5 dataset, our method improves the mAP50 from 38.8 to 45.2. Full article

Facial recognition technology has significantly advanced in recent years, with promising applications in fields ranging from security to consumer electronics. Its importance extends beyond convenience, offering enhanced security measures for sensitive areas and seamless user experiences in everyday devices. This study focuses on the development and validation of a facial recognition system utilizing a Haar cascade classifier and the AdaBoost machine learning algorithm. The system leverages characteristic facial features—distinct, measurable attributes used to identify and differentiate faces within images. A biometric facial recognition system was implemented on a Raspberry Pi microcomputer, capable of detecting and identifying faces using a self-contained reference image database. Verification involved selecting the similarity threshold, a critical factor influencing the balance between accuracy, security, and user experience in biometric systems. Testing under various environmental conditions, facial expressions, and user demographics confirmed the system’s accuracy and efficiency, achieving an average recognition time of 10.5 s under different lighting conditions, such as daylight, artificial light, and low-light scenarios. It is shown that the system’s accuracy and scalability can be enhanced through testing with larger databases, hardware upgrades like higher-resolution cameras, and advanced deep learning algorithms to address challenges such as extreme facial angles. Threshold optimization tests with six male participants revealed a value that effectively balances accuracy and efficiency. While the system performed effectively under controlled conditions, challenges such as biometric similarities and vulnerabilities to spoofing with printed photos underscore the need for additional security measures, such as thermal imaging. Potential applications include access control, surveillance, and statistical data collection, highlighting the system’s versatility and relevance. Full article

For surveillance video management in university laboratories, issues such as occlusion and low-resolution face capture often arise. Traditional face recognition algorithms are typically static and rely heavily on clear images, resulting in inaccurate recognition for low-resolution, small-sized faces. To address the challenges of occlusion and low-resolution person identification, this paper proposes a new face recognition framework by reconstructing Retinaface-Resnet and combining it with Quality-Adaptive Margin (adaface). Currently, although there are many target detection algorithms, they all require a large amount of data for training. However, datasets for low-resolution face detection are scarce, leading to poor detection performance of the models. This paper aims to solve Retinaface’s weak face recognition capability in low-resolution scenarios and its potential inaccuracies in face bounding box localization when faces are at extreme angles or partially occluded. To this end, Spatial Depth-wise Separable Convolutions are introduced. Retinaface-Resnet is designed for face detection and localization, while adaface is employed to address low-resolution face recognition by using feature norm approximation to estimate image quality and applying an adaptive margin function. Additionally, a multi-object tracking algorithm is used to solve the problem of moving occlusion. Experimental results demonstrate significant improvements, achieving an accuracy of 96.12% on the WiderFace dataset and a recognition accuracy of 84.36% in practical laboratory applications. Full article

This study proposes a novel log end face feature extraction and matching method based on Swin Transformer V2, aiming to address limitations in accuracy and speed faced by traditional deep learning models, like InceptionResNetV2 and Vision Transformer. Accurate log identification is crucial for forestry and wood supply chain management, especially given the growing reliance on timber imports to meet industrial demands in construction, furniture manufacturing, and paper production. Our dataset comprises images of coniferous timber, specifically Scots pine (Pinus sylvestris L.), reflecting its significance as an essential imported resource in China’s timber industry. By leveraging Swin Transformer V2 as the backbone, our method enhances feature extraction and achieves a significant accuracy improvement from 84.0% to 97.7% under random rotation angles while reducing the average matching time per log to 0.249 s. The model was evaluated under fixed and random rotation augmentations, and the results demonstrated Swin Transformer V2’s superior clustering ability, as confirmed by t-SNE visualization. Unlike InceptionResNetV2, the proposed model maintains high accuracy and efficiency even as the feature database size increases, making it suitable for large-scale applications. This approach provides a more accurate and efficient solution for log end-face recognition, supporting the development of high-throughput wood identification systems critical for forestry automation and the global timber trade. Full article

With the advent of the 5G era, high-precision localization based on mobile communication networks has become a research hotspot, playing an important role in indoor emergency rescue in shopping malls, smart factory management and tracking, as well as precision marketing. However, in complex environments, non-line-of-sight (NLOS) propagation reduces the measurement accuracy of 5G signals, causing large deviations in position solving. In order to obtain high-precision position information, it is necessary to recognize the propagation state of the signal before distance measurement or angle measurement. In this paper, we propose a dynamic multi-level optimization of comprehensive features (DMOCF) network model for line-of-sight (LOS)/NLOS identification. The DMOCF model improves the expression ability of the deep model by adding a res2 module to the time delay neural network (TDNN), so that fine-grained feature information such as weak reflections or noise in the signal can be deeply understood by the model, enabling the network to realize layer-level feature processing by adding Squeeze and Excitation (SE) blocks with adaptive weight adjustment for each layer. A mamba module with position coding is added to each layer to capture the local patterns of wireless signals under complex propagation phenomena by extracting local features, enabling the model to understand the evolution of signals over time in a deeper way. In addition, this paper proposes an improved sand cat search algorithm for network parameter search, which improves search efficiency and search accuracy. Overall, this new network architecture combines the capabilities of local feature extraction, global feature preservation, and time series modeling, resulting in superior performance in the 5G channel impulse response (CIR) signal classification task, improving the accuracy of the model and accurately identifying the key characteristics of multipath signal propagation. Experimental results show that the NLOS/LOS recognition method proposed in this paper has higher accuracy than other deep learning methods. Full article

With the digital transformation of the manufacturing industry, data monitoring and collecting in the manufacturing process become essential. Pointer meter reading recognition (PMRR) is a key element in data monitoring throughout the manufacturing process. However, existing PMRR methods have low accuracy and insufficient robustness due to issues such as blur, uneven illumination, tilt, and complex backgrounds in meter images. To address these challenges, we propose an end-to-end PMRR method based on a decoupled circle head detection algorithm (YOLOX-DC) and a Unet-like pure Transformer segmentation network (PM-SwinUnet). First, according to the characteristics of the pointer dial, the YOLOX-DC detection algorithm is designed based on the exceeding you only look once detector (YOLOX). The decoupled circle head of YOLOX-DC detects the pointer meter dial more accurately than the commonly used rectangular detection head. Second, the window multi-head attention of the PM-SwinUnet network enhances the feature extraction ability of pointer meter images and solves problems of missed scale detection and incomplete pointer segmentation. Additionally, the scale and pointer fitting module is introduced into the PM-SwinUnet to locate the accurate position of the scale and pointer. Finally, through the angle relationship between the pointer and the first two main scale lines, the pointer meter reading is accurately calculated by the improved angle method. Experimental results demonstrate the effectiveness and superiority of the proposed end-to-end method across three-pointer meter datasets. Furthermore, it provides a rapid and robust approach to the digital transformation of manufacturing systems. Full article

Over recent years, automated Human Activity Recognition (HAR) has been an area of concern for many researchers due to its widespread application in surveillance systems, healthcare environments, and many more. This has led researchers to develop coherent and robust systems that efficiently perform HAR. Although there have been many efficient systems developed to date, still, there are many issues to be addressed. There are several elements that contribute to the complexity of the task, making it more challenging to detect human activities, i.e., (i) poor lightning conditions; (ii) different viewing angles; (iii) intricate clothing styles; (iv) diverse activities with similar gestures; and (v) limited availability of large datasets. However, through effective feature extraction, we can develop resilient systems for higher accuracies. During feature extraction, we aim to extract unique key body points and full-body features that exhibit distinct attributes for each activity. Our proposed system introduces an innovative approach for the identification of human activity in outdoor and indoor settings by extracting effective spatio-temporal features, along with a Multi-Class Support Vector Machine, which enhances the model’s performance to accurately identify the activity classes. The experimental findings show that our model outperforms others in terms of classification, accuracy, and generalization, indicating its efficient analysis on benchmark datasets. Various performance metrics, including mean recognition accuracy, precision, F1 score, and recall assess the effectiveness of our model. The assessment findings show a remarkable recognition rate of around 88.61%, 87.33, 86.5%, and 81.25% on the BIT-Interaction dataset, UT-Interaction dataset, NTU RGB + D 120 dataset, and PKUMMD dataset, respectively. Full article