Object Size Research Articles

DBI-Attack: Dynamic Bi-Level Integrated Attack for Intensive Multi-Scale UAV Object Detection

Benefiting from the robust feature representation capability of convolutional neural networks (CNNs), the object detection technology of intelligent high-altitude UAV remote sensing has been developed rapidly. In this field, the adversarial examples (AEs) pose serious security risks and vulnerabilities to deep learning-based systems. Due to the limitation of object size, image degradation, and scene brightness, adding adversarial disturbances to small and dense objects is extremely challenging. To study the threat of AE for UAV object detection, a dynamic bi-level integrated attack (DBI-Attack) is proposed for intensive multi-scale UAV object detection. Firstly, we use the dynamic iterative attack (DIA) method to generate perturbation on the classification level by improving the momentum iterative fast gradient sign method (MIM). Secondly, the bi-level adversarial attack method (BAAM) is constructed to add global perturbation on the decision level for completing the white-box attack. Finally, the integrated black-box attack method (IBAM) is combined to realize the black-box mislabeling and fabrication attacks. We experiment on the real drone traffic vehicle detection datasets to better evaluate the attack effectiveness. The experimental results show that the proposed method can achieve mislabeling and fabrication attacks on the UAV object detectors in black-box conditions. Furthermore, the adversarial training is applied to improve the model robustness. This work aims to call more attention to the adversarial and defensive aspects of UAV target detection models.

A semi-parallel CNN-transformer fusion network for semantic change detection

Semantic change detection (SCD) can recognize the region and the type of changes in remote sensing images. Existing methods are either based on transformer or convolutional neural network (CNN), but due to the size of various ground objects is different, it is necessary to have global modeling ability and local information extraction ability at the same time. Therefore, in this paper we propose a fusion semantic change detection network (FSCD) with both global modeling ability and local information extraction ability by fusing transformer and CNN. A semi-parallel fusion block has also been proposed to construct FSCD. It can not only have global and local features in parallel, but also fuse them as deeply as serial. To better adaptively decide which mechanism is applied to which pixel, we design a self-attention and convolution selection module (ACSM). ACSM is a self-attention mechanism used to selectively combine transformer and CNN. Specifically, the importance of each mechanism is automatically obtained by learning. According to the importance, the mechanism suitable for a pixel is selected, which is better than using either mechanism alone. We evaluate the proposed FSCD on two datasets, and the proposed network has a significant improvement compared with the state-of-the-art network.

SHM to detect and characterize impact events in metallic aircraft structure

Aircraft structures are susceptible to foreign object impacts, which may occur during manufacturing, maintenance and in-service. Sizes of these impacted objects especially during in-service can range from a small rock to a large bird. Common ways to detect such impacts are based on flight / ground crew observations and reports leading to close examinations of structures using non-destructive evaluation (NDE) techniques. If undetected these impact damages can grow during service loading and may be detrimental to flight safety. Therefore, timely detection of any signs of impact damages are critical such that proper maintenance actions can be taken. The aim is to develop methodologies using Structural Health Monitoring (SHM) techniques to detect and characterize foreign object impact events. In this experiment, a cut-out of an aluminum panel measuring 31 x 26 inches from an out-of-service aircraft was used. The panel was instrumented with four Lead Zirconate Titanate (PZT) sensors from Acellent Inc., as well as, four Acoustic Emission (AE) sensors from Mistras Inc. Impedance and susceptance measurements were acquired to assess the proper functionality of the PZT sensors before and after the impact events. Both the PZTs and the AE sensors were directly connected to a digital oscilloscope, without any amplification and / or filtering for acquiring raw data during the impact events. An instrumented multi-use tapper was designed and developed to calibrate the system, as well as, to record the impulse (impact force and time) during the impact events. The acquired data were processed using physics-based and machine learning techniques to detect and characterize the impact events.

Soft robotic fingers with sensorized pre-charged finger pulp

Pneumatically driven soft actuators with sensors have been developing rapidly these years. They can perceive external stimulus and be applied to different scenarios. In this study, we present a novel soft robotic finger with sensorized finger pulp based on sealing a flexible fabric piezoresistive film called Velostat into a pre-charged air bag, which can perceive the contact force with an object based on changes in resistance value of the sensor. The soft sensor mimics human finger pulp and deforms passively according to the shape of objects during grasping, so that it can firmly contact with objects and as such improves the gripper’s grasping stability. Moreover, based on force feedback, the actuator can reduce or increase the input pressure to hold the object and control the contact force precisely. The sensor exhibits a sensitivity of up to 0.328 kPa−1 and can measure pressures ranging from 0 to over 10 kPa. The sensor’s measurement range and sensitivity can be pre-adjusted by regulating the pre-charged pressure during fabrication for different grasping tasks. The response/recovery time of the sensor is 80/60 ms on average. Experiments show that the finger with sensorized pulp can be applied for object softness and size detection, object transport minitoring as well as force control grasping. The proposed soft robotic finger has potential for applications in scenarios that require safe contact and closed-loop control.

A multiscale 3D network for lung nodule detection using flexible nodule modeling.

Lung cancer is the most common type of cancer. Detection of lung cancer at an early stage can reduce mortality rates. Pulmonary nodules may represent early cancer and can be identified through computed tomography (CT) scans. Malignant risk can be estimated based on attributes like size, shape, location, and density. Deep learning algorithms have achieved remarkable advancements in this domain compared to traditional machine learning methods. Nevertheless, many existing anchor-based deep learning algorithms exhibit sensitivity to predefined anchor-box configurations, necessitating manual adjustments to obtain optimal outcomes. Conversely, current anchor-free deep learning-based nodule detection methods normally adopt fixed-size nodule models like cubes or spheres. To address these technical challenges, we propose a multiscale 3D anchor-free deep learning network (M3N) for pulmonary nodule detection, leveraging adjustable nodule modeling (ANM). Within this framework, ANM empowers the representation of target objects in an anisotropic manner, with a novel point selection strategy (PSS) devised to accelerate the learning process of anisotropic representation. We further incorporate a composite loss function that combines the conventional L2 loss and cosine similarity loss, facilitating M3N to learn nodules' intensity distribution in three dimensions. Experiment results show that the M3N achieves 90.6% competitive performance metrics (CPM) with seven predefined false positives per scan on the LUNA 16 dataset. This performance appears to exceed that of other state-of-the-art deep learning-based networks reported in their respective publications. Individual test results also demonstrate that M3N excels in providing more accurate, adaptive bounding boxes surrounding the contours of target nodules. The newly developed nodule detection system reduces reliance on prior knowledge, such as the general size of objects in the dataset, thus it should enhance overall robustness and versatility. Distinct from traditional nodule modeling techniques, the ANM approach aligns more closely with the morphological characteristics of nodules. Time consumption and detection results demonstrate promising efficiency and accuracy which should be validated in clinical settings.

Do 14-17-month-old infants use iconic speech and gesture cues to interpret word meanings?a).

This experimental study investigated whether infants use iconicity in speech and gesture cues to interpret word meanings. Specifically, we tested infants' sensitivity to size sound symbolism and iconic gesture cues and asked whether combining these cues in a multimodal fashion would enhance infants' sensitivity in a superadditive manner. Thirty-six 14-17-month-old infants participated in a preferential looking task in which they heard a spoken nonword (e.g., "zudzud") while observing a small and large object (e.g., a small and large square). All infants were presented with an iconic cue for object size (small or large) (1) in the pitch of the spoken non-word (high vs low), (2) in gesture (small or large), or (3) congruently in pitch and gesture (e.g., a high pitch and small gesture indicating a small square). Infants did not show a preference for congruently sized objects in any iconic cue condition. Bayes factor analyses showed moderate to strong support for the null hypotheses. In conclusion, 14-17-month-old infants did not use iconic pitch cues, iconic gesture cues, or iconic multimodal cues (pitch and gesture) to associate speech sounds with their referents. These findings challenge theories that emphasize the role of iconicity in early language development.

Hand muscle synergy in chopstick use: effect of object size and weight

This study explains the role of muscle coordination in chopstick manipulation and investigates the effects of object width and weight on intrinsic and extrinsic hand muscle activity when picking up objects with chopsticks. Surface electromyography was used to measure the activity of the intrinsic and extrinsic hand muscles when picking up objects of varying widths and weights using chopsticks. The results revealed coordinated muscle activity patterns in the intrinsic and extrinsic hand muscles and coordination between them during chopstick manipulation. Object widths varying between 1 and 3 cm did not significantly affect muscle activity; however, object weight influenced muscle activity during both chopstick closing and object grasping, with greater muscle activity in the 40 g condition than in the 10 g condition. Intrinsic hand muscles were found to be involved in object grasping, regardless of object weight. These findings suggest that object weight should be considered when practicing picking up objects with chopsticks in scenarios resembling daily dining, to prevent excessive muscle activity during rehabilitation.

Radar Based Security System Alert

Abstract: Radar-based security systems are now a reliable option for both intruder detection and perimeter defence. This research presents a unique approach to integrate state-of-the-art signal processing methods to improve the performance of such systems. The suggested system tracks and detects moving objects inside an established region using radar technology. The system utilizes advanced processing of signals methods, such as fuzzy logic, neural networks, and wavelet analysis, to differentiate between everyday activities and possible security risks. The network of radar sensors that are placed strategically around the perimeter of the secured region makes up the radar-based security system. Every radar sensor keeps an eye on the environment and detects any moving objects that come within range. Wavelet analysis is used to analyze the collected signals in real-time and extract important properties, such as object velocity, direction, and size.

Real-time evaluation of object detection models across open world scenarios

Object detection models have been experiencing significant improvements over the years due to advancements in deep learning techniques, increased availability of large-scale annotated datasets, and computational resources. Different object detection models have varying levels of accuracy, speed, and robustness. With the increasing complexity and diversity of object detection models, it becomes a problem for researchers and practitioners to choose the most suitable model for their specific needs. This research paper outlines the escalating demand for robust comparison of object detection models in response to rapidly advancing technology. This evaluation helps in identifying the strengths and weaknesses of these models and selecting the most suitable one for a specific task. This highlights a significant challenge stemming from the lack of recent comparative studies on object detection models across various image qualities, object sizes, and training data sizes. The above challenges are tackled by a meticulous evaluation of three state-of-the-art object detection models: YOLO-v8, Faster R-CNN with ResNet 50 and 101 backbones, and End-to-End Object Detection Transformers (DETR) utilizing ResNet 50 and 101 backbones by employing a rigorous assessment framework encompassing mean Average Precision (mAP), accuracy, and inference speed. This study focuses on thoroughly examining how well the models perform across three different datasets: TACO, PlastOPol, and TACO 4.5. These datasets consist of open-world images captured in real-time from various locations. They include 1500, 2500, and 6500 images respectively, depicting real-world environments with varying lighting conditions and complex backgrounds. The results identify YOLOv8 as the superior model for high and medium-quality images, while Faster R-CNN performs better for low-quality images. However, DETR's accuracy falls short compared to other models. The paper fills a crucial gap in understanding model performance across varying image qualities and object sizes and helps in taking informed decisions in object detection systems.

On the use of scale distortion for visual humour

In contrast to verbal humour, visual humour remains a relatively underdeveloped area of research. In this exploratory study, we investigate whether scale incongruity – i.e., discrepancy between the expected and actual experience of the size of an object – can serve as a source of humour in the visual modality. We adapt a pre-existing visual data set of mundane scenes by altering the size of an individual object in each scene and collecting humorousness ratings from human annotators on the original and scale-distorted versions. Our analysis of these annotations reveals that scenes with distorted objects are perceived to be significantly funnier than the original images.

An Optimization Method for Lightweight Rock Classification Models: Transferred Rich Fine-Grained Knowledge.

Rock image classification represents a challenging fine-grained image classification task characterized by subtle differences among closely related rock categories. Current contrastive learning methods prevalently utilized in fine-grained image classification restrict the model's capacity to discern critical features contrastively from image pairs, and are typically too large for deployment on mobile devices used for in situ rock identification. In this work, we introduce an innovative and compact model generation framework anchored by the design of a Feature Positioning Comparison Network (FPCN). The FPCN facilitates interaction between feature vectors from localized regions within image pairs, capturing both shared and distinctive features. Further, it accommodates the variable scales of objects depicted in images, which correspond to differing quantities of inherent object information, directing the network's attention to additional contextual details based on object size variability. Leveraging knowledge distillation, the architecture is streamlined, with a focus on nuanced information at activation boundaries to master the precise fine-grained decision boundaries, thereby enhancing the small model's accuracy. Empirical evidence demonstrates that our proposed method based on FPCN improves the classification accuracy mobile lightweight models by nearly 2% while maintaining the same time and space consumption.

Contextual expectations shape the motor coding of movement kinematics during the prediction of observed actions: A TMS study

Contextual information may shape motor resonance and support intention understanding during observation of incomplete, ambiguous actions. It is unclear, however, whether this effect is contingent upon kinematics ambiguity or contextual information is continuously integrated with kinematics to predict the overarching action intention. Moreover, a differentiation between the motor mapping of the intention suggested by context or kinematics has not been clearly demonstrated. In a first action execution phase, 29 participants were asked to perform reaching-to-grasp movements towards big or small food objects with the intention to eat or to move; electromyography from the First Dorsal Interosseous (FDI) and Abductor Digiti Minimi (ADM) was recorded. Depending on object size, the intentions to eat or to move were differently implemented by a whole-hand or a precision grip kinematics, thus qualifying an action-muscle dissociation. Then, in a following action prediction task, the same participants were asked to observe an actor performing the same actions and to predict his/her intention while motor resonance was assessed for the same muscles. Of note, videos were interrupted at early or late action phases, and actions were embedded in contexts pointing toward an eating or a moving intention, congruently or incongruently with kinematics. We found greater involvement of the FDI or ADM in the execution of precision or whole-hand grips, respectively. Crucially, this pattern of activation was mirrored during observation of the same actions in congruent contexts, but it was cancelled out or reversed in the incongruent ones, either when videos were interrupted at either early or long phases of action deployment. Our results extend previous evidence by showing that contextual information shapes motor resonance not only under conditions of perceptual uncertainty but also when more informative kinematics is available.

Determinants of Maximum Magnetic Anomaly Detection Distance.

The maximum detection distance is usually the primary concern of magnetic anomaly detection (MAD). Intuition tells us that larger object size, stronger magnetization and finer measurement resolution guarantee a further detectable distance. However, the quantitative relationship between detection distance and the above determinants is seldom studied. In this work, unmanned aerial vehicle-based MAD field experiments are conducted on cargo vessels and NdFeB magnets as typical magnetic objects to give a set of visualized magnetic field flux density images. Isometric finite element models are established, calibrated and analyzed according to the experiment configuration. A maximum detectable distance map as a function of target size and measurement resolution is then obtained from parametric sweeping on an experimentally calibrated finite element analysis model. We find that the logarithm of detectable distance is positively proportional to the logarithm of object size while negatively proportional to the logarithm of resolution, within the ranges of 1 m~500 m and 1 pT~1 μT, respectively. A three-parameter empirical formula (namely distance-size-resolution logarithmic relationship) is firstly developed to determine the most economic sensor configuration for a given detection task, to estimate the maximum detection distance for a given magnetic sensor and object, or to evaluate minimum detectable object size at a given magnetic anomaly detection scenario.

Study of the dependence of accuracy in vehicles search on the size of the object using UAV images

Digital images are increasingly used to analyze different types of objects and their localization and classification. There are many areas for using this information; it is often employed for surveillance systems, automatic driving of vehicles, or exploration of new territories. At the same time, there are a fairly large number of neural networks that allow implementation of this functionality by training them using data sets of various types and classifications. Often, data sets created with the help of unmanned aerial vehicles are frequently used for research tasks. Such datasets allow the recognition of various types of objects without direct access to them, which allows safe exploration of different territories. The use of unmanned aerial vehicles is quite common nowadays, especially in the fields of photography and videography. Many photographers use unmanned aerial vehicles to take pictures of landscapes and use automatic tracking systems for movement. Automatic movement systems and object search systems are quite sensitive to the size of the object and the quality of the search algorithm. Because of the wide applicability of this task, as well as the small amount of initial data, the topic of our work is the study of the dependence of the accuracy of localization and classification of objects on their area in images obtained using unmanned aerial vehicles. The main subject of this study is the quality of neural networks that allow obtaining information about objects, as well as research by obtaining statistical data and a test set of data on the dependence of detection accuracy on the size of the object. The goal of this study was to obtain statistics on the accuracy of localization and classification depending on the size of the object and to determine the accuracy thresholds using the obtained statistics. The task of this study is to train common neural networks with an open architecture on a set of data obtained using unmanned aerial vehicles and to determine their characteristics, particularly the dependence of recognition accuracy on the size of the object. The expected result of the work is the threshold values of the size of the object, which are permissible for a sufficiently accurate classification and localization of objects, as well as the metrics of the quality of the work of the studied neural networks. Because of this work, conclusions are given that reflect the threshold values of object sizes, on which the recognition accuracy depends.

Construction workers' unsafe behavior detection through adaptive spatiotemporal sampling and optimized attention based video monitoring

In recent years, advances in construction site image analysis faced challenges, particularly in construction object detection and identifying unsafe actions. Challenges involve complex backgrounds, varying object sizes, and image quality. Existing methods address spatial and temporal features with attention mechanisms but often overlook adaptive sampling and channel-wise adjustments, missing potential spatiotemporal redundancies. This article introduces the Optimized Positioning (OP-Net) architectures and an attention-based spatiotemporal sampling approach. The OP module is introduced for object detection, which enhances channel relationships by leveraging global feature affinity associations. Additionally, we propose an innovative spatiotemporal sampling strategy that adapts to effectively identify unsafe actions in construction sites. We extensively evaluate the object detection task using the SODA dataset to showcase the efficacy and effectiveness of our approach. Furthermore, our unsafe action identification model is benchmarked on the CMA dataset, demonstrating its ability to achieve new state-of-the-art performance in accuracy while maintaining reasonable computational efficiency.

Evaluating Machine Learning-Based MRI Reconstruction Using Digital Image Quality Phantoms.

Quantitative and objective evaluation tools are essential for assessing the performance of machine learning (ML)-based magnetic resonance imaging (MRI) reconstruction methods. However, the commonly used fidelity metrics, such as mean squared error (MSE), structural similarity (SSIM), and peak signal-to-noise ratio (PSNR), often fail to capture fundamental and clinically relevant MR image quality aspects. To address this, we propose evaluation of ML-based MRI reconstruction using digital image quality phantoms and automated evaluation methods. Our phantoms are based upon the American College of Radiology (ACR) large physical phantom but created in k-space to simulate their MR images, and they can vary in object size, signal-to-noise ratio, resolution, and image contrast. Our evaluation pipeline incorporates evaluation metrics of geometric accuracy, intensity uniformity, percentage ghosting, sharpness, signal-to-noise ratio, resolution, and low-contrast detectability. We demonstrate the utility of our proposed pipeline by assessing an example ML-based reconstruction model across various training and testing scenarios. The performance results indicate that training data acquired with a lower undersampling factor and coils of larger anatomical coverage yield a better performing model. The comprehensive and standardized pipeline introduced in this study can help to facilitate a better understanding of the performance and guide future development and advancement of ML-based reconstruction algorithms.

FAGD-Net: Feature-Augmented Grasp Detection Network Based on Efficient Multi-Scale Attention and Fusion Mechanisms

Grasping robots always confront challenges such as uncertainties in object size, orientation, and type, necessitating effective feature augmentation to improve grasping detection performance. However, many prior studies inadequately emphasize grasp-related features, resulting in suboptimal grasping performance. To address this limitation, this paper proposes a new grasping approach termed the Feature-Augmented Grasp Detection Network (FAGD-Net). The proposed network incorporates two modules designed to enhance spatial information features and multi-scale features. Firstly, we introduce the Residual Efficient Multi-Scale Attention (Res-EMA) module, which effectively adjusts the importance of feature channels while preserving precise spatial information within those channels. Additionally, we present a Feature Fusion Pyramidal Module (FFPM) that serves as an intermediary between the encoder and decoder, effectively addressing potential oversights or losses of grasp-related features as the encoder network deepens. As a result, FAGD-Net achieved advanced levels of grasping accuracy, with 98.9% and 96.5% on the Cornell and Jacquard datasets, respectively. The grasp detection model was deployed on a physical robot for real-world grasping experiments, where we conducted a series of trials in diverse scenarios. In these experiments, we randomly selected various unknown household items and adversarial objects. Remarkably, we achieved high success rates, with a 95.0% success rate for single-object household items, 93.3% for multi-object scenarios, and 91.0% for cluttered scenes.

Real-time underwater object detection technology for complex underwater environments based on deep learning

Underwater object detection technology is crucial in many marine-related fields, including marine environmental monitoring, marine resource development, and marine ecological protection. However, this technology faces great challenges due to the poor quality of underwater optical images and the varying sizes of underwater objects. Therefore, we proposed an underwater optical detection network (UODN) based on the you only look once version 8 (YOLOv8) framework, which addresses these issues through the cross stage multi-branch (CSMB) module and large kernel spatial pyramid (LKSP) module. The aim of the CSMB module is to extract more features from underwater optical images to address the issue of poor image quality, while the LKSP module is designed to enhance the ability of the network to detect underwater objects of various scales. Furthermore, CSMBDarknet built by CSMB and LKSP can be used as the backbone of other underwater object detection algorithms for underwater feature extraction. Extensive experimental results on the underwater robot professional contest 2020 dataset revealed that the average precision (AP) of UODN increased by 1.0%, the AP50 of UODN increased by 1.1%, and the AP75 of UODN increased by 2.1% compared with those of the original YOLOv8s. Furthermore, UODN outperforms 12 state-of-the-art models on multiple underwater optical datasets, paving the way for future real-time and high-precision underwater object detection.

Land cover fraction mapping across global biomes with Landsat data, spatially generalized regression models and spectral-temporal metrics

Mapping land cover in highly heterogeneous landscapes is challenging, and classifications have inherent limitations where the spatial resolution of remotely sensed data exceeds the size of small objects. For example, classifications based on 30-m Landsat data do not capture urban or other heterogeneous environments well. This limitation may be overcome by quantifying the subpixel fractions of different land cover types. However, the selection process and transferability of models designed for subpixel land cover mapping across biomes is yet challenging. We asked to what extent (a) locally trained models can be used for sub-pixel land cover fraction estimates in other biomes, and (b) training data from different regions can be combined into spatially generalized models to quantify fractions across global biomes. We applied machine learning regression-based fraction mapping to quantify land cover fractions of 18 regions in five biomes using Landsat data from 2022. We used spectral-temporal metrics to incorporate intra-annual temporal information and compared the performance of local, spatially transferred, and spatially generalized models. Local models performed best when applied to their respective sites (average mean absolute error, MAE, 9–18%), and also well when transferred to other sites within the same biome, but not consistently so for out-of-biome sites. However, spatially generalized models that combined input data from many sites worked very well when analyzing sites in many different biomes, and their MAE values were only slightly higher than those of the respective local models. A weighted training data selection approach, preferring training data with a lower spectral distance to the image data to be predicted, further enhanced the performance of generalized models. Our results suggest that spatially generalized regression-based fraction models can support multi-class sub-pixel fraction estimates based on medium-resolution satellite images globally. Such products would have great value for environmental monitoring in heterogeneous environments and where land cover varies along spatial or temporal gradients.

AeroDetectNet: a lightweight, high-precision network for enhanced detection of small objects in aerial remote sensing imagery

Object detection in remote sensing imagery exhibits difficulties due to complex backgrounds, diverse object scales, and intricate spatial context relationships. Motivated by the problems mentioned above, this paper introduces AeroDetectNet, a novel lightweight and high-precision object detection network custom-designed for aerial remote sensing scenarios, building upon the YOLOv7-tiny algorithm. It enhances performance through four key improvements: the normalized Wasserstein distance for consistent object size sensitivity, the Involution module for reduced background noise, a self-designed RCS-biformer module for better spatial context interpretation, and a self-designed WF-CoT SPPCSP feature pyramid for improved feature map weighting and context capture. Ablation studies conducted on a hybrid dataset composed of three open-source remote sensing datasets (including NWPU VHR-10 remote sensing images, RSOD remote sensing images, and VisDrone UAV images) have demonstrated the effectiveness of four improvements specifically for small-size object detection. Visualizations through Grad-CAM further demonstrate AeroDetectNet’s capacity to extract and focus on key object features. Upon individual testing across three open-source datasets, AeroDetectNet has successfully demonstrated its ability to identify objects in images with a smaller pixel area. Through experimental comparisons with other related studies, the AeroDetectNet achieved a competitive mAP while maintaining fewer model parameters, highlighting its highly accurate and lightweight properties.