Object Size Research Articles

Single Object Tracking in Satellite Videos Based on Feature Enhancement and Multi-Level Matching Strategy

Despite significant advancements in remote sensing object tracking (RSOT) in recent years, achieving accurate and continuous tracking of tiny-sized targets remains a challenging task due to similar object interference and other related issues. In this paper, from the perspective of feature enhancement and a better feature matching strategy, we present a tracker SiamTM specifically designed for RSOT, which is mainly based on a new target information enhancement (TIE) module and a multi-level matching strategy. First, we propose a TIE module to address the challenge of tiny object sizes in satellite videos. The proposed TIE module goes along two spatial directions to capture orientation and position-aware information, respectively, while capturing inter-channel information at the global 2D image level. The TIE module enables the network to extract discriminative features of the targets more effectively from satellite images. Furthermore, we introduce a multi-level matching (MM) module that is better suited for satellite video targets. The MM module firstly embeds the target feature map after ROI Align into each position of the search region feature map to obtain a preliminary response map. Subsequently, the preliminary response map and the template region feature map are subjected to the Depth-wise Cross Correlation operation to get a more refined response map. Through this coarse-to-fine approach, the tracker obtains a response map with a more accurate position, which lays a good foundation for the prediction operation of the subsequent sub-networks. We conducted extensive experiments on two large satellite video single-object tracking datasets: SatSOT and SV248S. Without bells and whistles, the proposed tracker SiamTM achieved competitive results on both datasets while running at real-time speed.

A quantitative CBCT pipeline based on 2D antiscatter grid and grid-based scatter sampling for image-guided radiation therapy.

Quantitative accuracy is critical for expanding the role of cone beam CT (CBCT) imaging from target localization to quantitative treatment monitoring and plan adaptations in radiation therapy. Despite advances in CBCT image quality improvement methods, quantitative accuracy gap between CBCT and multi-detector CT (MDCT) remains. In this work, a physics-driven approach was investigated that combined robust scatter rejection, raw data correction and iterative image reconstruction to further improve CBCT image quality and quantitative accuracy, referred to as quantitative CBCT (qCBCT). QCBCT approach includes tungsten 2D antiscatter grid hardware, residual scatter correction with grid-based scatter sampling, image lag, and beam hardening correction for offset detector geometry linac-mounted CBCT. Images were reconstructed with iterative image reconstruction to reduce image noise. qCBCT was evaluated using a variety of phantoms to investigate the effect of object size and its composition on image quality, and image quality was benchmarked against clinical CBCT and gold standard MDCT images used for treatment planning. QCBCT provided statistically significant improvement in CT number accuracy and reduced image artifacts when compared to clinical CBCT images. When compared to gold standard MDCT, mean HU errors in qCBCT and clinical CBCT were 17±9 and 38±29 HU, respectively. Magnitude of phantom size dependent HU variations were comparable between MDCT and qCBCT images. With iterative reconstruction, contrast-to-noise ratio improved by 25% when compared to clinical CBCT protocols. Combination of novel scatter suppression techniques and other data correction methods in qCBCT provided CT number accuracy comparable to gold standard MDCT used for treatment planning. This approach may potentially improve CBCT's promise in fulfilling the tasks that demand high quantitative accuracy, such as online dose calculations and treatment response assessment, in image guided radiation therapy.

Deep Slice-Crossed Network With Local Weighted Loss for Brain Metastases Segmentation

Cognitive deficits occur in up to 90% of patients with brain metastases, can be caused by the tumor itself and whole brain radiation therapy. Latest treatment options to reduce cognitive decline require high-precision metastases segmentation. This asks existing metastases segmentation models to be further improved as their limited ability to micro lesion perception. To reduce the segmentation error on brain tissues and improve the accuracy on micro metastases, a deep slice-crossed network and a local weighted loss are proposed, respectively. The slice-crossed network contains a gated feature fusion module where the "gate" is learned individually from the slice-crossed label to roughly localize metastases. Then in tumor regions, this "gate" allows intra-slice information to pass for metastases enhancement, while in other regions, it allows inter-slice information to pass for vascular screening. The local weighted loss is proposed to address the inconsistency in size of intra-class objects. Unlike existing weighting approaches, it calculates the weight of each pixel from its neighbourhood pixel distribution to enhance the learning of pixels within small tumours, which can be applied to common losses. To evaluate them, two datasets containing 1000 and 368 patients, respectively, were collated. Experimental results demonstrate that the proposed method perform satisfactorily compared with state-of-the-art brain metastasis segmentation models.

Design and manufacture of a soft robot with dual-interaction in virtual reality

This paper examines the design and fabrication of a soft robot that can connect to a virtual reality environment. This study's primary objective is to utilize these technologies concurrently and demonstrate their applicability in various applications, particularly rehabilitation. Therefore, the process of designing and modeling the soft robot is carried out, and an applied model is created using a 3D printer and silicon material, which is then installed on gloves. Using Unity software, a virtual reality environment is created in which programs, commands, and Arduino processors control the movements of the soft robot, allowing the user to move and pick up an object in a real environment while wearing gloves, and to adjust the amount of pressure and angle of its motion based on the size of each virtual object. During the system evaluation phase, a delay in the performance and reaction time of the soft robot installed on the gloves is observed. This delay is reduced by modifying the programming structure, resulting in optimal system functionality. This capability is used to create proper mobility conditions and rehabilitation for the majority of patients with wrist injuries resulting from strokes and accidents, and it may be effective in accelerating patients' recoveries.

Simulation of Foreign Object Detection Using Passive Inductive Sensors in a Wireless Charging System for Electric Vehicles

During wireless charging of the traction battery of electrically powered vehicles, the active area between the ground and vehicle assemblies must be monitored for inductive power transfer. If metallic foreign objects enter this area, they interact with the magnetic field and can heat up strongly, and thus become a potential source of hazard. To detect such foreign objects, measurements based on passive inductive sensors have already been carried out in advance. However, a large number of factors influence the detectability of metallic foreign objects, such as the characteristics of the magnetic field of the ground assembly coil, the size, shape, position, orientation, and material composition of the foreign objects, or the design of the sensor coils. The related practical testing effort can be reduced if the characteristics of the charging system and the foreign object detection system can be simulated. Therefore, simulation models were developed within the scope of this work and validated with the help of practical measurements. These models were used in the next step to analyze new test arrangements that had not yet been investigated by measurement. In the simulations described here, precision in the range of 1 mV could be achieved. Cumulatively, many influencing factors can be easily investigated, and results can be generated in a largely automated manner and typically in a wider variety than with practical measurements.

Cloud reflection modelling for impact flashes on Jupiter

Aims. We investigate the optical characteristics of flashes caused by the impact of metre- to decametre-sized outer Solar System objects on Jupiter, and the contributions of reflected light from surface clouds at visible wavelengths, in order to estimate more accurate bulk parameters, such as the luminous energy of the flash, the kinetic energy, the mass, and the size of the impact object. Methods. Using the results of recent reflectivity studies of the Jovian surface, we developed a cloud reflection model that calculates the contribution of the reflected light relative to that directly from the flash. We compared the apparent luminous energy of the previously reported flashes with the expected cloud reflection contributions to obtain their revised bulk parameters. Results. We find that cloud reflection contributions can make up to 200% of the flux directly from the flash, and thus can be the most significant uncertainty in the measurement of the bulk parameters. The reflection contributions strongly depend on wavelength. With our cloud reflection correction, we obtained revised bulk parameters for the previously reported flashes. Conclusions. Our cloud reflection correction enables us to better understand the properties of objects impacting Jupiter and is crucial for ongoing detailed investigations using high-sensitivity and multi-wavelength observation systems, such as PONCOTS. It will also be useful for understanding other optical transients in Jupiter’s upper atmosphere, such as the recently discovered sprite-like events.

A Novel Artificial Visual System for Motion Direction Detection with Completely Modeled Retinal Direction-Selective Pathway

Some fundamental visual features have been found to be fully extracted before reaching the cerebral cortex. We focus on direction-selective ganglion cells (DSGCs), which exist at the terminal end of the retinal pathway, at the forefront of the visual system. By utilizing a layered pathway composed of various relevant cells in the early stage of the retina, DSGCs can extract multiple motion directions occurring in the visual field. However, despite a considerable amount of comprehensive research (from cells to structures), a definitive conclusion explaining the specific details of the underlying mechanisms has not been reached. In this paper, leveraging some important conclusions from neuroscience research, we propose a complete quantified model for the retinal motion direction selection pathway and elucidate the global motion direction information acquisition mechanism from DSGCs to the cortex using a simple spiking neural mechanism. This mechanism is referred to as the artificial visual system (AVS). We conduct extensive testing, including one million sets of two-dimensional eight-directional binary object motion instances with 10 different object sizes and random object shapes. We also evaluate AVS’s noise resistance and generalization performance by introducing random static and dynamic noises. Furthermore, to thoroughly validate AVS’s efficiency, we compare its performance with two state-of-the-art deep learning algorithms (LeNet-5 and EfficientNetB0) in all tests. The experimental results demonstrate that due to its highly biomimetic design and characteristics, AVS exhibits outstanding performance in motion direction detection. Additionally, AVS possesses biomimetic computing advantages in terms of hardware implementation, learning difficulty, and parameter quantity.

Swin-YOLO for Concealed Object Detection in Millimeter Wave Images

Concealed object detection in millimeter wave (MMW) images has gained significant attention in the realm of public safety, primarily due to its distinctive advantages of non-hazardous and non-contact operation. However, this undertaking confronts substantial challenges in practical applications, owing to the inherent limitations of low imaging resolution, small concealed object size, intricate environmental noise, and the need for real-time performance. In this study, we propose Swin-YOLO, an innovative single-stage detection model built upon transformer layers. Our approach encompasses several key contributions. Firstly, the integration of Local Perception Swin Transform Layers (LPST Layers) enhanced the network’s capability to acquire contextual information and local awareness. Secondly, we introduced a novel feature fusion layer and a specialized prediction head for detecting small targets, effectively leveraging the network’s shallow feature information. Lastly, a coordinate attention (CA) module was seamlessly incorporated between the neck network and the detection head, augmenting the network’s sensitivity towards critical regions of small objects. To validate the efficacy and feasibility of our proposed method, we created a new MMW dataset containing a large number of small concealed objects and conducted comprehensive experiments to evaluate the effectiveness of overall and partial improvements, as well as computational efficiency. The results demonstrated a remarkable 4.7% improvement in the mean Average Precision (mAP) for Swin-YOLO compared with the YOLOv5 baseline. Moreover, when compared with other enhanced transformer-based models, Swin-YOLO exhibited a superior accuracy and the fastest inference speed. The proposed model showcases enhanced performance and holds promise for advancing the capabilities of real-world applications in public safety domains.

Adaptive Adjacent Layer Feature Fusion for Object Detection in Remote Sensing Images

Object detection in remote sensing images faces the challenges of a complex background, large object size variations, and high inter-class similarity. To address these problems, we propose an adaptive adjacent layer feature fusion (AALFF) method, which is developed on the basis of RTMDet. Specifically, the AALFF method incorporates an adjacent layer feature fusion enhancement (ALFFE) module, designed to capture high-level semantic information and accurately locate object spatial positions. ALFFE also effectively preserves small objects by fusing adjacent layer features and employs involution to aggregate contextual information in a wide spatial range for object essential features extraction in complex backgrounds. Additionally, the adaptive spatial feature fusion (ASFF) module is introduced to guide the network to select and fuse the crucial features to improve the adaptability to objects with different sizes. The proposed method achieves mean average precision (mAP) values of 77.1%, 88.9%, and 95.7% on the DIOR, HRRSD, and NWPU VHR-10 datasets, respectively. Notably, our approach achieves mAP75 values of 60.8% and 79.0% on the DIOR and HRRSD datasets, respectively, surpassing the state-of-the-art performance on the DIOR dataset.

Design and Implementation of a Compliant Gripper for Form Closure of Diverse Objects

This research presents a comprehensive study on the design and implementation of a flexible robotic gripper. Conventional grippers utilized in articulated robotic arms are often limited in their capabilities, being restricted to specific tasks or fixed object sizes. While soft grippers are a viable option, they have limitations in terms of grasping objects across a wide range and providing complete coverage. In this study, a novel compressible gripper is developed to enable safe and secure grasping of objects with varying sizes and shapes within a wide range. The gripper features a grasping area measuring 14 cm × 6 cm, allowing complete coverage of objects within this surface area. The current prototype with 7 cm of compressibility demonstrates the ability to grasp objects with a width difference of 7 cm with a maximum thickness of 15 cm, enabling manipulation of objects with varying widths, as defined by user-programmable parameters. The functionality of the gripper is based on the compressibility of the 3D-printed thermoplastic polyurethane (TPU) material. The flexible part of the gripper can be easily interchanged, offering versatility by accommodating different thicknesses without the need to replace the entire gripper mechanism. The gripper system operates using an open-loop control system, enhancing user-friendliness. Experimental evaluation of the gripper involved the creation and analysis of a CAD model followed by the fabrication of a prototype. The prototype exhibited exceptional performance in grasping objects of diverse sizes, shapes, and textures, demonstrating the effectiveness of the developed soft gripper system. The scalability of the soft gripper enables seamless integration with various types of articulated robotic arms, while the maximum weight limit for objects will be defined based on the robotic arms’ limitations. The research findings highlight the promising capabilities of the compressible gripper in enhancing the versatility and efficiency of robotic grasping systems, offering a significant contribution to the field of robotics.

Parallel wavelet networks incorporating modality adaptation for hyperspectral image super-resolution

Hyperspectral images (HSIs) have been applied to a wide range of areas thanks to their high spectral resolutions. However, spatial resolutions of HSIs are inevitably compromised due to hardware limits. While HSIs can be enhanced by the super resolution techniques, there may be still spectral distortions and spatial blurriness in the reconstructed high-resolution HSIs (HR-HSIs). Therefore, this paper proposes a Parallel Framework based on Modality-Adaptive Wavelet Networks (PF-MAWN) to reconstruct HR-HSISs by effectively fusing low-resolution HSIs (LR-HSIs) and high-resolution multispectral images (HR-MSIs). To avoid spectral artifacts, multiple MAWNs work in parallel and fuse HR-MSIs/LR-HSIs in groups of highly correlated adjacent bands. Each MAWN incorporates wavelet transform (WT) deeply into convolutional neural networks (CNN) by layers of Multidomain Preprocessing Blocks (MPBs), Multimodal Adaptation Blocks (MABs) and Multilevel Connection Blocks (MCBs). Within each MAB, there are two Feature Alignment Modules (FAMs) based on self-attention to radiometrically match HR-MSIs/LR-HSIs and an Information Control Module (ICM) based on Gaussian-weighted attention to control textural details in HR-HSIs. The framework exhibits good extensibility and generalizability on spectral images of arbitrary bands and diversified scales. Experimental results show that PF-MAWN can reconstruct high-quality HR-HSIs with different sizes of objects, ranging from microscopic to macroscopic. It outperforms several state-of-the-art methods on four public datasets of histopathology tissues, everyday stuff and remote sensing objects.

Hierarchical Feature Association and Global Correction Network for Change Detection

Optical satellite image change detection has attracted extensive research due to its comprehensive application in earth observation. Recently, deep learning (DL)-based methods have become dominant in change detection due to their outstanding performance. Remote sensing (RS) images contain different sizes of ground objects, so the information at different scales is crucial for change detection. However, the existing DL-based methods only employ summation or concatenation to aggregate several layers of features, lacking the semantic association of different layers. On the other hand, the UNet-like backbone is favored by deep learning algorithms, but the gradual downscaling and upscaling operation in the mainstream UNet-like backbone has the problem of misalignment, which further affects the accuracy of change detection. In this paper, we innovatively propose a hierarchical feature association and global correction network (HFA-GCN) for change detection. Specifically, a hierarchical feature association module is meticulously designed to model the correlation relationship among different scale features due to the redundant but complementary information among them. Moreover, a global correction module on Transformer is proposed to alleviate the feature misalignment in the UNet-like backbone, which, through feature reuse, extracts global information to reduce false alarms and missed alarms. Experiments were conducted on several publicly available databases, and the experimental results show the proposed method is superior to the existing state-of-the-art change detection models.

Effect of Slice Thickness Variation on Spatial Resolution of CT-Scan Device Images at the Bali Mandara Hospital Radiology Installation

Aims: Slice thickness variation analysis on spatial resolution on CT-scan images by determining CNR and SNR values, by making ROI on object and background images. ROI size of objects and backgrounds can be observed with several criteria based on the provisions of the ACR program, so that optimization of CT-Scan image quality can be achieved.
 Place and Duration of Study: The Radiology Installation in Bali Mandara Hospital, between 23 March to 20 May 2022.
 Methodology: Data collection begins with preparing the equipment used, namely CT-Scan, head holder, phantom, and computer. After that, the head holder is placed in the middle of the gantry and then the phantom is placed with the guidance of the laser beam (alignment system). Set the axial beam on the vertical line of section 1 and then set the coronal beam on the horizontal line of section 2. After that, setting the type of examination using the adult head protocol and setting the scan parameters to 80 kV, 100 mAs, and slice thicknesses 1, 2, 3, 4, 6, and 8 mm. After the preparation is done then the next exposure to the phantom. Then the research data in the form of phantom images will be seen as a whole to program on the computer monitor. Measurements were made 5 times for each slice thickness. Next is to analyze the CNR of digital phantom image data on a computer. So that it can be determined object ROI, background ROI, object mean, background mean, and standard deviation of the background.
 Results: The results showed that the thicker the Slice Thickness, the higher the CNR and SNR values. The highest CNR value occurred in air medium (13.70) and the lowest occurred in Nylon (0.86). Likewise, the highest SNR value occurred in air medium (13.84) and the lowest in Nylon (0.99). The highest CNR and SNR values ​​occurred in air medium and the lowest occurred in Nylon. 
 Conclusion: Slice Thickness variations affect the spatial resolution of CT-Scan images as indicated by the CNR and SNR values however, the CNR and SNR values ​​are still above 1.0 for adult head examination and above 0.7 for children's head examination. So that the quality of CT-Scan images at Bali Mandara Hospital is still very good.

CT Number Accuracy and Association With Object Size: A Phantom Study Comparing Energy-Integrating Detector CT and Deep Silicon Photon-Counting Detector CT.

BACKGROUND. Variable beam hardening based on patient size causes variation in CT numbers for energy-integrating detector (EID) CT. Photon-counting detector (PCD) CT more accurately determines effective beam energy, potentially improving CT number reliability. OBJECTIVE. The purpose of the present study was to compare EID CT and deep silicon PCD CT in terms of both the effect of changes in object size on CT number and the overall accuracy of CT numbers. METHODS. A phantom with polyethylene rings of varying sizes (mimicking patient sizes) as well as inserts of different materials was scanned on an EID CT scanner in single-energy (SE) mode (120-kV images) and in rapid-kilovoltage-switching dual-energy (DE) mode (70-keV images) and on a prototype deep silicon PCD CT scanner (70-keV images). ROIs were placed to measure the CT numbers of the materials. Slopes of CT number as a function of object size were computed. Materials' ideal CT number at 70 keV was computed using the National Institute of Standards and Technology XCOM Photon Cross Sections Database. The root mean square error (RMSE) between measured and ideal numbers was calculated across object sizes. RESULTS. Slope (expressed as Hounsfield units per centimeter) was significantly closer to zero (i.e., less variation in CT number as a function of size) for PCD CT than for SE EID CT for air (1.2 vs 2.4 HU/cm), water (-0.3 vs -1.0 HU/cm), iodine (-1.1 vs -4.5 HU/cm), and bone (-2.5 vs -10.1 HU/cm) and for PCD CT than for DE EID CT for air (1.2 vs 2.8 HU/cm), water (-0.3 vs -1.0 HU/cm), polystyrene (-0.2 vs -0.9 HU/cm), iodine (-1.1 vs -1.9 HU/cm), and bone (-2.5 vs -6.2 HU/cm) (p < .05). For all tested materials, PCD CT had the smallest RMSE, indicating CT numbers closest to ideal numbers; specifically, RMSE (expressed as Hounsfield units) for SE EID CT, DE EID CT, and PCD CT was 32, 44, and 17 HU for air; 7, 8, and 3 HU for water; 9, 10, and 4 HU for polystyrene; 31, 37, and 13 HU for iodine; and 69, 81, and 20 HU for bone, respectively. CONCLUSION. For numerous materials, deep silicon PCD CT, in comparison with SE EID CT and DE EID CT, showed lower CT number variability as a function of size and CT numbers closer to ideal numbers. CLINICAL IMPACT. Greater reliability of CT numbers for PCD CT is important given the dependence of diagnostic pathways on CT numbers.

Morphological-based Analyses for Parameterizing Symmetry in Radio Galaxies

The morphology of radio galaxies can provide significant clues to describe the formation and evolution of galaxies in the Universe. Here, we aim to extract the morphological parameters of radio galaxies and define symmetry criteria as some of the essential factors of their shape explanations. We employed 67 radio galaxies, which include Fanaroff–Riley type 1 and type 2 galaxies, and their radio images from the FIRST and LoTSS surveys. We developed an automatic segmentation process to extract morphological properties such as the size of objects, eccentricity, and orientation of segmented regions from data sets. Using a maximum likelihood estimator, we show that the distributions of sizes follow a power-law function with exponents of −0.39 ± 0.06 and −0.55 ± 0.05 for the FIRST and LoTSS data, respectively. We found that type 2 radio galaxies have slightly lower eccentricities than type 1. We studied the relationships between size, eccentricity, and redshift in scatter plots. The size of galaxies (kpc2) demonstrates gently growing trends with increasing eccentricity in their scatter plots. We discussed the possible effect of the redshifts of the galaxies on this result. Depending on the number of segmented regions, we defined symmetry criteria based on proximity to the center of a galaxy in the optical band, eccentricity, orientation, and the quarter (q) of appearance in the image. We found that the mean symmetry obtained for two segmented regions that mainly emerged in two quarters via the condition of has a higher value than those obtained for other cases.

A Benchmark for Maritime Object Detection with Centernet on an Improved Dataset, ABOships-PLUS

Object detection from waterborne imagery is an essential aspect in maritime traffic management, navigation safety and coastal security. Building efficient autonomous systems, which can take decisions in critical situations, requires an integration of complex object detectors in real time. Object detectors trained on generic datasets often give unsatisfactory results in complex scenarios like the maritime environment, since only a fraction of their images contain maritime vessels. Publicly available domain-specific datasets are scarce, and they are limited in the number of images and annotations. Compared to object detection in applications such as autonomous vehicles, maritime vessel detection is considerably reduced in computer vision research. This creates a deficit in exhaustive benchmarking studies for maritime detection datasets. To bridge this gap, we relabel the ABOships dataset and benchmark a state-of-the-art center-based detector, Centernet, on the newly relabeled dataset, ABOships-PLUS. We explore its performance under different feature extractors, and investigate the effect of object size and inter-class variation on detection accuracy. The reported benchmarking illustrates that the ABOships-PLUS dataset is adequate to use in supervised domain adaptation. The experimental results show that Centernet with DLA (Deep Layer Aggregation) as a feature extractor achieved the highest accuracy in detecting maritime objects overall (with mean average precision 74.4%).

Лингвистические средства выражения ускорения и замедления времени в различных типах художественной речи (на материале произведений англо-американских авторов)

The paper examines linguistic means that express slowed and accelerated flow of time in the novels by American and British authors. The aim of the research is to demonstrate the fact that such types of literary speech as description and narration can effectively express different speed of time in combination with linguistic means. The scientific novelty of the work consists in determining the atypical meaning of Future Simple and Present Progressive in the context that reveals the character’s perception of time, as well as in identifying the specific contents of narration that can contribute to the transmission of time acceleration by linguistic means: changes in the size of objects, their rotational movement, illusion of counter motion, the character’s rapid movement in space. The results of the study showed that polysyndeton can create opposite effects: slow down (description of subjects) and accelerate time (narration about planned actions). It was also found that Present Progressive, if used unconventionally in the narration about events of the distant past, renders the slowing down of time.

Identification and detection of microplastic particles in marine environment by using improved faster R–CNN model

Microplastics refer to plastic particles measuring less than 5 mm, which has led to serious environmental problem and the detection of these tiny particles is crucial for understanding the corresponding distribution and impact on the marine environment. In this paper, an improved faster region-based convolutional neural network (R–CNN) model was developed for the identification and detection of microplastic particles. In the proposed model, the residual network-50 (ResNet-50) is employed as the backbone with the replacement of the traditional one to enhance the feature extraction capability and the feature pyramid networks (FPN) module is introduced together for solving the multi-scale target detection. By using the improved Faster R–CNN model, the network model performance is enhanced where the average confidence of detecting unique microplastic particles in the marine environment reaches as high as 99%. Moreover, the microparticles mixture was bounded precisely via the predicted bounding boxes without missing detection and wrong detection. In this way, the successful identification of polystyrene microplastic particles from the particles suspension with similar shapes but various conditions of backgrounds, brightness, distributions and object sizes, was achieved by employing the proposed improved Faster R–CNN model, enabling the accurate detection of microplastic particles in marine environment.

Small object detection in UAV image based on improved YOLOv5

Less effective information is obtained by the object detection network, due to the small size of the detection object in the entire image, the complex background, and the dense object in unmanned aerial vehicle (UAV) images. In response to the difficulties encountered, a small object detection method in UAV images is proposed as an improved YOLOv5-based algorithm in this paper. First, the space-to-depth(SPD) conv module is introduced into the basic feature extraction network, to improve significant loss of image information during downsampling. Then, various attention mechanisms are added, to intensify the acquisition of regions of interest in UAV images. Finally, the multiscale detection module is improved, to enhance the network's ability to detect small objects in UAV images. By conducting experiments on the VisDrone-DET2019 dataset, the test results of the established model show. The improved algorithm achieved a Mean Average Precision (mAP) of 41.8%, which is 7.8% better than the baseline network. In addition, the detection performance is better than most current mainstream target detection algorithms and is of some practical value.

Ellipse Detection with Applications of Convolutional Neural Network in Industrial Images

Ellipse detection has a very wide range of applications in the field of industrial production, especially in the geometric detection of metallurgical hinge pins. However, the factors in industrial images, such as small object size and incomplete ellipse in the image boundary, bring challenges to ellipse detection, which cannot be solved by existing methods. This paper proposes a method for ellipse detection in industrial images, which utilizes the extended proposal operation to prevent the loss of ellipse rotation angle features during ellipse regression. Moreover, the Gaussian angle distance conforming to the ellipse axioms is adopted and combined with smooth L1 loss as the ellipse regression loss function to enhance the prediction accuracy of the ellipse rotation angle. The effectiveness of the proposed method is demonstrated on the hinge pins dataset, with experiment results showing an AP* of 80.93% and indicating superior detection performance compared to other methods. It is thus suitable for engineering applications and can provide visual guidance for the precise measurement of ellipse-like mechanical parts.