Object Size Research Articles

Target-aware cross-modality unsupervised domain adaptation for vestibular schwannoma and cochlea segmentation.

There is growing interest in research on segmentation for the vestibular schwannoma (VS) and cochlea using high-resolution T2 (hrT2) imaging over contrast-enhanced T1 (ceT1) imaging due to the contrast agent side effects. However, the hrT2 imaging remains a problem of insufficient annotated data, which is fatal for building more robust segmentation models. To address the issue, recent studies have adopted unsupervised domain adaptation approaches that translate ceT1 images to hrT2 images. However, previous studies did not consider the size and visual characteristics of the target objects, such as VS and cochlea, during image translation. Specifically, those works simply performed normalization on the entire image without considering its significant impact on the quality of the translated images. These approaches tend to erase the small target objects, making it difficult to preserve the structure of these objects when generating pseudo-target images. Furthermore, they may also struggle to accurately reflect the unique style of the target objects within the images. Therefore, we propose a target-aware unsupervised domain adaptation framework, designed for translating target objects, each tailored to their unique visual characteristics and size using target-aware normalization. We demonstrate the superiority of the proposed framework on a publicly available challenge dataset. Codes are available at https://github.com/Bokyeong-Kang/TANQ .

Scale-Adaptive Salience Supervision and Dynamic Token Filtering for Small Object Detection in Remote Sensing Images

Abstract Recently, DETR-like detectors, which have shown remarkable performance in general object detection, face limitations when dealing with remote sensing images primarily containing small objects. Mainstream two-stage DETR-like models employ a pipeline that selects and processes a small portion of informative tokens, which enhances performance but also shows a high dependency on token selection. The current static token selection strategies lead to inconsistencies between the static selection criteria and dynamic token updates. Additionally, in remote sensing images, the limited information available for small objects and their inherent sensitivity to pixel shifts further degrade detection performance. To address this, we propose Scale-Adaptive Salience DETR (SAS DETR), a two-stage DETR-like method. SAS DETR incorporates dynamic token filtering, which uses a global threshold predictor to determine the token filtering ratio for each layer of the encoder. This approach selects an appropriate filtering ratio for different network layers while maintaining consistency between the foreground confidence map and token updates. Furthermore, we introduce a novel scale-adaptive salience supervision mechanism that adaptively scales the salience computation area based on object size, ensuring the model more effectively supervises small objects and utilizes the information within tokens without compromising the detection performance for objects of other sizes. Finally, we employ Scale-Adaptive Intersection over Union to reduce the impact of pixel shifts on small objects. With these improvements, our proposed SAS DETR achieves 25.2% AP on the AI-TOD-V2 dataset with 24 training epochs and 50.4% AP on the COCO 2017 dataset with 12 training epochs.

RUSNet: Robust fish segmentation in underwater videos based on adaptive selection of optical flow

Fish segmentation in underwater videos can be used to accurately determine the silhouette size of fish objects, which provides key information for fish population monitoring and fishery resources survey. Some researchers have utilized underwater optical flow to improve the fish segmentation accuracy of underwater videos. However, the underwater optical flow is not evaluated and screen in existing works, and its predictions are easily disturbed by motion of non-fish. Therefore, in this paper, by analyzing underwater optical flow data, we propose a robust underwater segmentation network, RUSNet, with adaptive screening and fusion of input information. First, to enhance the robustness of the segmentation model to low-quality optical flow inputs, a global optical flow quality evaluation module is proposed for evaluating and aligning the underwater optical flow. Second, a decoder is designed by roughly localizing the fish object and then applying the proposed multidimension attention (MDA) module to iteratively recover the rough localization map from the spatial and edge dimensions of the fish. Finally, a multioutput selective fusion method is proposed in the testing stage, in which the mean absolute error (MAE) of the prediction using a single input is compared with that obtained using multisource input. Then, the information with the highest confidence is selected for predictive fusion, which facilitates the acquisition of the ultimate underwater fish segmentation results. To verify the effectiveness of the proposed model, we trained and evaluated it using a publicly available joint underwater video dataset and a separate DeepFish public dataset. Compared with the advanced underwater fish segmentation model, the proposed model has greater robustness to low-quality background optical flow in the DeepFish dataset, with the mean pixel accuracy (mPA) and mean intersection over union (mIoU) values reaching 98.77% and 97.65%, respectively. On the joint dataset, the mPA and mIoU of the proposed model are 92.61% and 90.12%, respectively, which are 0.72% and 1.21% higher than those of the advanced underwater video object segmentation model MSGNet. The results indicate that the proposed model can adaptively select the input and accurately segment fish in complex underwater scenes, which provides an effective solution for investigating fishery resources.

STEM-18. A HIGH THROUGHPUT NEUROSPHERE BASED COLONY FORMATION ASSAY TO VALIDATE DRUG AND RADIATION EFFECTIVENESS IN PATIENT DERIVED XENOLINES

Abstract Glioblastoma (GBM) is a primary malignant brain tumor developed from normal astrocytes. GBM patients’ median survival is less than 2 years despite safe surgical resection, fractionated conventional radiation, and temozolomide (TMZ) therapy. Few currently used drugs are successful in GBM although there are many FDA approved blood-brain barrier (BBB) penetrant drugs, for other uses that have not been screened in GBM. We utilized a panel of 734 of these BBB penetrant drugs in different GBM patient derived xenolines (PDX) along with acquired radio-resistant derivatives (-RT) including X1441, JX14P, JX14P-RT, JX39P, JX39P-RT and normal human astrocytes (NHA) as control and screened for viability with Cell Titer-Glo. We identified multiple candidate drugs that preferentially killed GBM PDX cells at 10 µM relative to NHA control. As we used stem cell media (serum free) in our drug screening study, it indicates that the potent drugs are also effective in killing cancer stem cells or brain tumor initiating cells. These drugs had targets ranging from PDGFR, DNA/RNA synthesis, as well as other kinase inhibitors. We developed a high throughput (HT), neurosphere based colony formation assay (CFA) in 96-well plate to screen multiple drugs and doses in combination with radiation. For visualizing the colonies, we use a Cytation 5 imager at 4X resolution and count the colonies based on parameters including object size, width, intensity and others. We plan on validating our drug screen and HT-CFA assay with high throughput PamChip kinase activity assays and with in vivo orthotopic tumor PDX models.

Assessment of object-based indices to identify convective organization

Abstract. The emerging field of convective organization has attracted significant attention due to its potential implications for weather and climate. Numerous indices have been developed to identify organization of convection, serving as essential tools for advancing our understanding in this area. Because of the large number of these indices, many results on convective organization are still uncertain, and different studies have shown diverging results. The present analysis studies and compares nine object-based indices in order to evaluate their ability to quantify organization. The analysis begins by establishing a set of criteria expected for convective organization and subsequently subjecting the indices to assessment against these benchmarks. The criteria are grouped into three categories. The first category tests the robustness of the indices against noise. The second category evaluates their sensitivity to the size and position of the convective objects. The third category assesses their dependency on the specific characteristics of the dataset in use. Among the indices scrutinized, none fulfill all the desired conditions, and some conditions are only marginally satisfied. Therefore, we developed a new index, called the Organization Index based on Distance and Relative Area (OIDRA), as an example of a well-behaving index. The unmet conditions and differences between indices can explain the discord between different organization studies. The results come down to a guideline that will help to advance our description of deep convective organization.

Visual Processing by Hierarchical and Dynamic Multiplexing.

The complexity of natural environments requires highly flexible mechanisms for adaptive processing of single and multiple stimuli. Neuronal oscillations could be an ideal candidate for implementing such flexibility in neural systems. Here, we present a framework for structuring attention-guided processing of complex visual scenes in humans, based on multiplexing and phase coding schemes. Importantly, we suggest that the dynamic fluctuations of excitability vary rapidly in terms of magnitude, frequency and wave-form over time, i.e., they are not necessarily sinusoidal or sustained oscillations. Different elements of single objects would be processed within a single cycle (burst) of alpha activity (7-14 Hz), allowing for the formation of coherent object representations while separating multiple objects across multiple cycles. Each element of an object would be processed separately in time-expressed as different gamma band bursts (>30 Hz)-along the alpha phase. Since the processing capacity per alpha cycle is limited, an inverse relationship between object resolution and size of attentional spotlight ensures independence of the proposed mechanism from absolute object complexity. Frequency and wave-shape of those fluctuations would depend on the nature of the object that is processed and on cognitive demands. Multiple objects would further be organized along the phase of slower fluctuations (e.g., theta), potentially driven by saccades. Complex scene processing, involving covert attention and eye movements, would therefore be associated with multiple frequency changes in the alpha and lower frequency range. This framework embraces the idea of a hierarchical organization of visual processing, independent of environmental temporal dynamics.

Point spread function engineering enable resolution enhanced imaging for Interferenceless coded aperture correlation holography

The point spread function (PSF) of an optical system could characterize the resolving ability of the whole optical system for point light sources. Therefore, the imaging performance of the system could be significantly improved by regulating and optimizing the PSF. In this paper, we innovatively propose a single-exposure hologram resolution enhanced cross-correlation (RECC) method for Interferenceless coded aperture holography(I-COACH) system, circumventing the necessity to obtain the point spread hologram (PSH) of an ideal point object. The RECC method firstly acquires an approximate image of a large-size point object by Lucy-Richardson (LR) algorithm in lens imaging mode, and takes it as a PSF to acquire a PSH with ideal size of the I-COACH system by LR algorithm again, and finally acquires a reconstructed image by the single-exposure hologram RECC method. In the RECC method, the approximate ideal PSHs at different axial positions of the system are acquired by offline operation, therefore, it has a high imaging temporal resolution, and the imaging transverse resolution is not affected by the size of the point objects at the time of recording the PSH, which provides a high imaging signal-to-noise ratio and stable resolution. The proposed method provides powerful technical support for further extending the application field of the I-COACH system, and provides technical reference for other incoherent imaging.

Frequency dependence of reflections on radar landmarks

Purpose. Reducing the dispersion of radar reflections from local objects, with multi-frequency sensing, to solve the problem of orientation by radar reflections from objects. Methodology. Reflections from local objects in the entire frequency range of the radar station (RS) at the same radio engineering position were measured by three independent radars of the same type on different days and at different antenna elevations. The deviation of the radar stations in the position did not exceed 500 meters. Coordinates (azimuth, range) of reflections of several separate local objects were allocated for each radar. The average values of reflections from local objects and their dispersion in the frequency range were calculated. Using various algorithms, individual frequencies were sampled and the reflected signals were averaged at these frequencies. The decrease in the dispersion of the reflected signal from the number of frequencies at which reflections were measured and from the algorithm for selecting these frequencies was investigated. Findings. Averaging the values of reflections from local objects for several frequencies leads to a decrease in dispersion and, as a result, to a more accurate correspondence of the reflected signal level to the geometric size of the local object. The variance decreases most rapidly for a small number of frequencies selected for averaging when selecting frequencies located in an interval of at least 1% relative to each other. Originality. To solve the problem of orientation based on radar reflections from local objects, it is necessary to identify the landmarks selected on a digital terrain model. Due to the fact that local objects (hills) are a collection of many reflectors falling into the allowed volume of the radar, with different levels of reflections and random phases, there may not be radar reflection from a local object at a certain frequency, or it may be very small. In order to unambiguously identify all landmarks, measurements must be carried out at several frequencies. The work has established how many frequencies measurements should be performed at and on what principle these frequencies should be selected. Practical value. The advent of digital terrain models made it possible to solve the problem of terrain orientation by comparing radar reflections from local objects with reflection models based on digital terrain maps. Radar reflection models use mathematical expectations of reflection values, unlike real reflections, which have random deviations in signal levels depending on the operating frequency. Reducing the variance of these deviations increases the accuracy of identifying characteristic local objects (landmarks) used to orient the radar in the absence of data from satellite navigation systems.

Demystifying the effect of receptive field size in U-Net models for medical image segmentation.

Medical image segmentation is a critical task in healthcare applications, and U-Nets have demonstrated promising results in this domain. We delve into the understudied aspect of receptive field (RF) size and its impact on the U-Net and attention U-Net architectures used for medical imaging segmentation. We explore several critical elements including the relationship among RF size, characteristics of the region of interest, and model performance, as well as the balance between RF size and computational costs for U-Net and attention U-Net methods for different datasets. We also propose a mathematical notation for representing the theoretical receptive field (TRF) of a given layer in a network and propose two new metrics, namely, the effective receptive field (ERF) rate and the object rate, to quantify the fraction of significantly contributing pixels within the ERF against the TRF area and assessing the relative size of the segmentation object compared with the TRF size, respectively. The results demonstrate that there exists an optimal TRF size that successfully strikes a balance between capturing a wider global context and maintaining computational efficiency, thereby optimizing model performance. Interestingly, a distinct correlation is observed between the data complexity and the required TRF size; segmentation based solely on contrast achieved peak performance even with smaller TRF sizes, whereas more complex segmentation tasks necessitated larger TRFs. Attention U-Net models consistently outperformed their U-Net counterparts, highlighting the value of attention mechanisms regardless of TRF size. These insights present an invaluable resource for developing more efficient U-Net-based architectures for medical imaging and pave the way for future exploration of other segmentation architectures. A tool is also developed, which calculates the TRF for a U-Net (and attention U-Net) model and also suggests an appropriate TRF size for a given model and dataset.

Criteria for objects suitable for reconstruction from holograms and diffraction patterns

In this study, quantitative criteria for reconstruction of objects from their hologram and diffraction patterns, and in particular for the phase objects in digital holography, are derived. The criteria that allow distinguishing the hologram and diffraction pattern are outlined. Gabor derived his criterion for objects suitable for holography based on the condition that the background in the reconstructed object’s distribution should be nearly flat so that its intensity contrast does not exceed 0.05. According to Gabor, an opaque object is suitable for holographic reconstruction if it occupies no more than 1% of the imaged area, and a phase-shifting object cannot be reconstructed in principle. We revisit these criteria and show that both amplitude-only and phase-only objects can be reconstructed when the object occupies less than 1% of the total illuminated area. In addition, a simplified derivation of the criteria is provided that is based on Parseval’s theorem. It is shown that for objects (including amplitude-only and phase-only) reconstructed from their holograms and the twin image treated as noise, a signal-to-noise ratio of 10 or higher can be achieved provided the object occupies less than 0.5% of the total illuminated area. When a hologram is reconstructed by applying iterative algorithms, the requirement for the object size is much more generous and identical to that applied in coherent diffraction imaging: any type of object (amplitude-only, phase-only, or amplitude-and-phase mixed properties) is suitable for holography when the object’s size in each dimension is less than half of the probed region’s extent (or the field of view).

CSSDet: small object detection via cross-scale feature enhancement on drone-view images

ABSTRACT Object detection on drone-view images is vital for applications like intelligent transportation, abnormal behavior detection, and urban surveillance. However, the diverse perspectives and altitudes from which Unmanned Aerial Vehicle (UAV) capture scenes pose challenges due to multi-scale differences and small object sizes. To address this, we propose CSSDet, emphasizing multi-scale object detection and small object enhancement. CSSDet integrates an additional detection head for small objects and employs a bidirectional weighted feature pyramid network for effective cross-scale feature fusion, enhancing global perceptual capability. Additionally, we introduce coordinate attention and Involution modules to mitigate information loss. Experiments are conducted on two publicly available datasets, VisDrone and UAVDT. Quantitative results from the experiments indicate that the proposed CSSDet achieves optimal performance across a wide range of evaluation metrics, with particular strengths demonstrated in multi-scale object detection and small-size object detection. Qualitative results showcase its remarkable effectiveness in various aspects of object detection, including small object detection, objects at different scales and categories, multi-perspective scenarios, and complex scenes. The substantial experimental findings confirm that CSSDet attains comprehensive object detection performance on drone-view images. The source code is available at https://github.com/Gui-Cheng/CSSDet.

An end-to-end model of active electrosensation.

1Weakly electric fish localize and identify objects by sensing distortions in a self-generated electric field. Fish can determine the resistance and capacitance of an object, for example, even though the field distortions being sensed are small and highly-dependent on object distance and size. Here we construct a model of the responses of the fish's electroreceptors on the basis of experimental data, and we develop a model of the electric fields generated by the fish and the distortions due to objects of different resistances and capacitances. This provides us with an accurate and efficient method for generating large artificial data sets simulating fish interacting with a wide variety of objects. Using these sets, we train an artificial neural network (ANN), representing brain areas downstream of electroreceptors, to extract the 3D location, size, and electrical properties of objects. The model performs best if the ANN operates in two stages: first estimating object distance and size and then using this information to extract electrical properties. This suggests a specific form of modularity in the electrosensory system that can be tested experimentally and highlights the potential of end-to-end modeling for studies of sensory processing.

EEG Responses to the Number of Objects in Partially Occluded and Uncovered Scenes.

Perceptual completion is ubiquitous when estimating properties such as the shape, size, or number of objects in partially occluded scenes. Behavioral experiments showed that the number of hidden objects is underestimated in partially occluded scenes compared with an estimation based on the density of visible objects and the amount of occlusion. It is still unknown at which processing level this (under)estimation of the number of hidden objects occurs. We studied this question using a passive viewing task in which observers viewed a game board that was initially partially occluded and later was uncovered to reveal its hidden parts. We simultaneously measured the electroencephalographic responses to the partially occluded board presentation and its uncovering. We hypothesized that if the underestimation is a result of early sensory processing, it would be observed in the activities of P1 and N1, whereas if it is because of higher level processes such as expectancy, it would be reflected in P3 activities. Our data showed that P1 amplitude increased with numerosity in both occluded and uncovered states, indicating a link between P1 and simple stimulus features. The N1 amplitude was highest when both the initially visible and uncovered areas of the board were completely filled with game pieces, suggesting that the N1 component is sensitive to the overall Gestalt. Finally, we observed that P3 activity was reduced when the density of game pieces in the uncovered parts matched the initially visible parts, implying a relationship between the P3 component and expectation mismatch. Overall, our results suggest that inferences about the number of hidden items are reflected in high-level processing.

Towards Autonomous Retail Stocking and Picking: Methods Enabling Robust Vacuum-Based Robotic Manipulation in Densely Packed Environments.

With the advent of robotics and artificial intelligence, the potential for automating tasks within human-centric environments has increased significantly. This is particularly relevant in the retail sector where the demand for efficient operations and the shortage of labor drive the need for rapid advancements in robot-based technologies. Densely packed retail shelves pose unique challenges for robotic manipulation and detection due to limited space and diverse object shapes. Vacuum-based grasping technologies offer a promising solution but face challenges with object shape adaptability. The study proposes a framework for robotic grasping in retail environments, an adaptive vacuum-based grasping solution, and a new evaluation metric-termed grasp shear force resilience-for measuring the effectiveness and stability of the grasp during manipulation. The metric provides insights into how retail objects behave under different manipulation scenarios, allowing for better assessment and optimization of robotic grasping performance. The study's findings demonstrate the adaptive suction cups' ability to successfully handle a wide range of object shapes and sizes, which, in some cases, overcome commercially available solutions, particularly in adaptability. Additionally, the grasp shear force resilience metric highlights the effects of the manipulation process, such as in shear force and shake, on the manipulated object. This offers insights into its interaction with different vacuum cup grasping solutions in retail picking and restocking scenarios.

Size sound symbolism in mothers' speech to their infants.

Six-month-olds infer object size based on pitch: they map high-pitched vowels onto smaller objects and low-pitched vowels onto larger objects (Peña et al., 2011). The 'sound symbolism bootstrapping hypothesis' (Imai & Kita, 2014) proposes that this may support understanding of word-meaning correspondences; by drawing on iconic pairings between linguistic cues and corresponding referents (e.g., higher pitch for smaller objects), infants develop understanding of word-referent associations. Here we analyse mother-child interactions, testing whether sound-symbolic pitch-size correspondences occur in infant-directed speech. 40 dyads engaged in (semi-)naturalistic interaction around picture books containing images of toys of varying sizes. We compare mothers' pitch when referring to small versus large toys, analysing i) actual size (bigger vs. smaller toys); ii) relative toy size congruence (i.e., congruent vs. incongruent with real-world expectation); and iii) transparency of the test paradigm as regards its focus on size contrastiveness (non-transparent, moderately transparent, highly transparent) to observe the nuances of size sound symbolism in infant-directed speech.

SODD-YOLOv8: an insulator defect detection algorithm based on feature enhancement and variable row convolution

Abstract To address the detection challenges in defective images of insulators in transmission lines, including tiny object size, significant scale variations, a wide variety of defects, and complex background interference. In this study, an improved insulator defect detection algorithm is proposed, based on the YOLOv8s framework and combining feature enhancement and deformable convolution techniques. Firstly, to address the image feature distortion problem caused by aerial photography, a deformable convolutional feature extraction module (DCFEM) is introduced, which is designed to enhance the model’s ability to adapt to the local geometric deformation, so as to effectively recover the distorted feature information in the image. Moreover, to enhance the detection ability of the model for small objects, a small object feature enhancement module is designed, which adopts an efficient multi-scale attention mechanism, and aims to enhance the feature extraction ability of small objects, improve the sensitivity to small-size defects, and improve the detection accuracy. Eventually, to optimize the computational efficiency of the model, the average pooling-sparse convolution-batch normalization (BN) module is proposed. This module combines average pooling, sparse convolution and BN techniques to achieve a lightweight model while maintaining a high level of feature extraction capability. Experimental results on the China power line insulator dataset show that the improved model achieves a 4.3 percentage point improvement in the mAP metric compared to YOLOv8s, and the number of parameters in the model is reduced by 10%. The proposed scheme not only improves the accuracy and efficiency of defect detection, but also reduces the demand for computational resources, thus providing a more reliable and efficient solution for insulator defect detection in practical applications.

Comparative Evaluation of Kanchnar Guggulu and Cystogrit Diamond in Stanagranthi (Fibroadenoma Breast)

Fibroadenoma is a benign breast tumor common in young women, affecting their daily life. Modern medicine often recommends surgical intervention, which can have physical and emotional consequences. The study explores alternative treatments through Ayurveda, which offers non-invasive and cost-effective options. This study investigates the comparative efficacy of Kanchnar Guggulu and Cystogrit Diamond in the management of Stanagranthi (Fibroadenoma breast). Sixty patients were randomly divided into two groups: Group A (treated with Kanchnar Guggulu) and Group B (treated with Cystogrit Diamond). The study assessed subjective (pain, consistency) and objective (size, number of lumps) parameters before and after a 90-day treatment period. The study aims to provide a safe, effective alternative to surgery for managing fibroadenoma, with Ayurveda offering promising non-invasive treatments. Results showed a significant reduction in pain for both groups, with Group B demonstrating a 93.24% improvement compared to Group A's 83.93%. In terms of lump size reduction, Group B also outperformed Group A, with a 42.86% decrease versus 29.85%, respectively. The average effect on consistency was higher in Group B (31.58%) compared to Group A (25%). The findings suggest that Cystogrit Diamond is more effective than Kanchnar Guggulu in managing Stanagranthi, providing a potential non-invasive, cost-effective alternative to surgery for fibroadenoma patients.

All-Optical Control of Bubble and Skyrmion Breathing.

Controlling the dynamics of topologically protected spin objects by all-optical means promises enormous potential for future spintronic applications. Excitation of bubbles and skyrmions in ferrimagnetic [Fe(0.35 nm)/Gd(0.40 nm)]_{160} multilayers by ultrashort laser pulses leads to a periodic modulation of the core diameter of these spin objects, the so-called breathing mode. We demonstrate versatile amplitude and phase control of this breathing using a double excitation scheme, where the observed dynamics is controlled by the excitation delay. We gain insight into both the timescale on which the breathing mode is launched and the role of the spin object size on the dynamics. Our results demonstrate that ultrafast optical excitation allows for precise tuning of the spin dynamics of trivial and nontrivial spin objects, showing a possible control strategy in device applications.

Modelling-assisted geometrical optimization of colloidal quantum color convertor based pixels fabricated by dielectrophoretic directed assembly

Hypothesis: Building competitive color conversion pixels for microdisplays made of semiconductor nanocrystals requires reaching a deposition thickness high enough to absorb all the blue light from the backlight unit. In the case of dielectrophoretic directed assembly of such nanocrystals, modeling and simulations may help understand what the intrinsic limitations of the process are, and may be used to propose new assembly routes.Experiments: A theoretical model of dielectrophoretic interactions between polarizable nano-spheres and an electrostatically patterned substrate has been developed. Monte Carlo simulations have been run using this model to rationalize the effects of parameters driving the dielectrophoretic directed assembly and to find optimal deposition conditions for reaching a maximal thickness of nanocrystal pixels. Experiments with CdSe quantum plates and with alumina spheres embedding quantum plates (micro-pearls) have been carried out and compared to the model.Findings: Modeling and simulations reveal that the directed assembly of semiconductor nanocrystals is limited essentially by the small object size, which sets the maximum dielectrophoretic force they can undergo. They indicate that using larger objects should allow reaching unprecedented assembly heights, but will induce lateral extension of the assembly. This trade-off has been illustrated with diagrams in the parameter space and confirmed experimentally with micro-pearls.

Lightweight Detection Algorithm for Air-to-Ground Threat Object

Abstract In air-to-ground threat object detection, the typically strong concealment, small size, and high movement speed of threat objects often lead to challenges such as missed detections, low accuracy, and an excessive number of network parameters. An efficient lightweight detection algorithm for air-to-ground threat objects is proposed, built upon the YOLOv7-tiny framework. To address the issue of large network parameters, lightweight modules ELAN-GS and Slim-Neck are employed. Additionally, to refine the detection precision for small objects, the NWD metric and NWD loss function are introduced. Results from experiments on the DOTA dataset highlight that, compared with the original network, the proposed algorithm achieves a 33.2% reduction in parameters and a 26.3% reduction in computational cost. The small object accuracy, mAPs@0.5:0.95, is 23.7%, which is a 4.4% enhancement relative to the original network. The significant reduction in parameters and computational cost, along with the substantial improvement in small object detection accuracy, demonstrate the effectiveness of the proposed algorithm.