Target Object Research Articles

Alignable kernel network

To enhance the adaptability and performance of Convolutional Neural Networks (CNN), we present an adaptable mechanism called Alignable Kernel (AliK) unit, which dynamically adjusts the receptive field (RF) dimensions of a model in response to varying stimuli. The branches of AliK unit are integrated through a novel align transformation softmax attention, incorporating prior knowledge through rank ordering constraints. The attention weightings across the branches establish the effective RF scales, leveraged by neurons in the fusion layer. This mechanism is inspired by neuroscientific observations indicating that the RF dimensions of neurons in the visual cortex vary with the stimulus, a feature often overlooked in CNN architectures. By aggregating successive AliK ensembles, we develop a deep network architecture named the Alignable Kernel Network (AliKNet). AliKNet with interdisciplinary design improves the network’s performance and interpretability by taking direct inspiration from the structure and function of human neural systems, especially the visual cortex. Empirical evaluations in the domains of image classification and semantic segmentation have demonstrated that AliKNet excels over numerous state-of-the-art architectures, achieving this without increasing model complexity. Furthermore, we demonstrate that AliKNet can identify target objects across various scales, confirming their ability to dynamically adapt their RF sizes in response to the input data.

Shape-Guided Detection: A joint network combining object detection and underwater image enhancement together

Most of the existing underwater image object detection methods involve pre-processing, such as using underwater image enhancement, to improve the accuracy of object detection. However, pre-processing methods are designed to improve the subjective perception of the human eye, which does not necessarily improve the object detection performance and consumes a large amount of computational resources. Therefore, in this paper, we creatively combine these two tasks and propose a Shape-Guided Detection network (SGD) to simultaneously optimize underwater image enhancement and object detection. In the SGD network, we innovatively incorporate the prior shape features as a learnable module embedded in it to fully explore the shape characteristics and structural details of the target object. To ensure that the prior knowledge can be effectively fused into the global network structure, we design a Shape Prior Enhancement module, which aims to realize the deep integration of the prior information with the local details. In order to optimize the stability of model training and enhance its convergence performance, a dual strategy of explicit and implicit constraints is ingeniously proposed in our method. We conduct extensive experiments on public datasets and the results show that the combination of our method with different detectors significantly improves the performance. The object detection performance reaches up to 0.491 mAP for optical images and 0.576 mAP for sonar images, and improves the preprocessing speed by 0.1 s.

Comparative Analysis of Classification and Segmentation Performance of Different Dog Breeds Based on Mask R-CNN

The classification of different image data is a common task in real life, which has been achieved as early as convolutional neural networks. With the emergence of R-CNN (Region CNN), simple classification tasks began to develop towards object detection tasks, followed by various segmentation tasks. After Faster R-CNN [4], Mask R-CNN [3] achieved instance segmentation while detecting objects. This algorithm model needs to distinguish different target objects in the image panel and learn a large number of characteristics to represent the details of each target object. In dog data images, there are differences and similarities in the characteristics of dog species, such as fur color, texture, and body shape. So in this article, the author selected and created two different dog datasets, representing targets with significant differences and similarities in features. Compare and analyze the performance of Mask R-CNN object detection and instance segmentation using these two dog datasets.

Algae-based self-driven microrobot for efficient removal of nanoplastics from water environment

Nanoplastics (NPs) have the characteristics of various species, wide distribution, low concentration and difficult degradation. In recent years, the researches of NPs have mainly focused on its toxicity, origin and migration, and the quantitative detection and removal technology of NPs is an urgent technical problem to be solved. Here, an active biohybrid microrobots (DPA-algae robots, diphenolic acids-alage robots) was designed and developed for dynamic removal of NPs in water environment. Firstly, diphenolic acids were functionalized on algae to realize the specific adsorption of nanoplastics through hydrophobic force, electrostatic attraction and van der Waals force between diphenolic acids and nanoplastic particles. Secondly, to realize the rapid identification of NPs, the functionalized algae were assembled into microrobots through rapid click chemistry reaction, thus the self-propulsion ability of algae can be utilized to accelerate the identification and enrichment of target objects. The removal rate of nanoplastics by microalgae robots has increased to 83.1 % at the concentration of 0.125 mg/mL within 2 h compared with the unmodified algae, which is a very significant improvement. In addition, the removal rate was 84.1 % to 87.7 % in different media, so the dynamic removal of NPs is expected to be applied into the filtration process of waterworks by preparing a large number of algae-based microrobots. This multi-functional microrobot is cost-effective and biocompatible, providing a new solution to the global “plastic crisis”.

Semantic ghost imaging based on semantic coding

To address the challenge of reconstructing high-quality images under low sampling during the Ghost Imaging (GI) procedure, we propose a semantic GI method based on semantic encoding. Through training, the obtained continuous weights of the convolution kernels are served as the speckle patterns. A processing module and a Recurrent Neural Network (RNN) are then employed to decode and reconstruct the images. Both the speckle patterns and their corresponding bucket values are optimized, so that the semantic information of the target object can be better extracted. With different gated recurrent units (GRU) layers for the target objects in different datasets, the feasibility of the proposed GI method is validated by the numerical simulations and experiments on the simple target objects (the handwriting digits “0” to “9”) and more complex target objects (“NUPT”). The results show that the target objects (the handwriting digits) can be reconstructed with higher quality even at a lower sampling rate of 1.28%. Additionally, the proposed method has applicability for more complex objects (such as “NUPT”) in real applications. In comparison with those results by using traditional ghost imaging (TGI), deep convolution auto-encoder network (DCAN), and RNN-based GI (GI-RNN), the proposed GI method shows a better performance in terms of the quality of reconstructed images and the training time, that is, the proposed method can have both good reconstructed image quality and less training time. By introducing the concept of semantic communication into GI, the proposed method provides a new idea for the model-based GI.

Visual search and real-image similarity: An empirical assessment through the lens of deep learning.

The ability to predict how efficiently a person finds an object in the environment is a crucial goal of attention research. Central to this issue are the similarity principles initially proposed by Duncan and Humphreys, which outline how the similarity between target and distractor objects (TD) and between distractor objects themselves (DD) affect search efficiency. However, the search principles lack direct quantitative support from an ecological perspective, being a summary approximation of a wide range of lab-based results poorly generalisable to real-world scenarios. This study exploits deep convolutional neural networks to predict human search efficiency from computational estimates of similarity between objects populating, potentially, any visual scene. Our results provide ecological evidence supporting the similarity principles: search performance continuously varies across tasks and conditions and improves with decreasing TD similarity and increasing DD similarity. Furthermore, our results reveal a crucial dissociation: TD and DD similarities mainly operate at two distinct layers of the network: DD similarity at the intermediate layers of coarse object features and TD similarity at the final layers of complex features used for classification. This suggests that these different similarities exert their major effects at two distinct perceptual levels and demonstrates our methodology's potential to offer insights into the depth of visual processing on which the search relies. By combining computational techniques with visual search principles, this approach aligns with modern trends in other research areas and fulfils longstanding demands for more ecologically valid research in the field of visual search.

Label-Free Exosome Analysis by Surface-Enhanced Raman Scattering Spectroscopy with Laser-Ablated Silver Nanoparticle Substrate.

Early diagnostics of breast cancer is crucial to reduce the risk of cancer metastasis and late relapse. Exosome, which contains distinct information of its origin, can be the target object as a liquid biopsy. However, its low sensitivity and inadequate diagnostic tools interfere with the point-of-care testing (POCT) of the exosome. Recently, Surface-enhanced Raman Scattering (SERS) spectroscopy, which amplifies the Raman scattering, has been proved as a promising tool for exosome detection. However, the fabrication process of SERS probe or substrate is still inefficient and far from large-scale production. This study proposes rapid and label-free detection of breast cancer-derived exosomes by statistical analysis of SERS spectra using silver-nanoparticle-based SERS substrate fabricated by selective laser ablation and melting (SLAM). Employing silver nanowires and optimizing laser process parameters enable rapid and low-energy fabrication of SERS substrate. The functionalities including sensitivity, reproducibility, stability, and renewability are evaluated using rhodamine 6G as a probe molecule. Then, the feasibility of POCT is examined by the statistical analysis of SERS spectra of exosomes from malignant breast cancer cells and non-tumorigenic breast epithelial cells. The presented framework is anticipated to be utilized in other biomedical applications, facilitating cost-effective and large-scale production performance.

Predicting the higher heating value of products through solid yield in torrefaction process

Torrefaction is a widely employed method to upgrade biomass for energy purpose. The higher heating value (HHV) serves as a vital indicator for assessing the energy potential of biomass. Nevertheless, HHV measurement is a time-consuming and costly process. HHV of raw biomass, torrefied biomass, and biochar has been extensively estimated using the results of elemental analysis and/or proximate analysis. However, these data must be repeatedly measured for each target object. To reduce the efforts required for the measurement of each object, this study proposes, for the first time, the use of solid yield (the most easily obtainable data) to predict the HHV of torrefied biomass.A total of 215 sets of data were retrieved from different biomass types undergoing various torrefaction technologies. Artificial neural network (ANN) and hierarchical linear regression (HLR) were employed to develop prediction models. Solid yield demonstrated a clear correlation to HHV for each individual biomass; however, diverse biomass types exhibited significant variation. Therefore, the results of elemental analysis of raw feedstocks were incorporated to account for variations in feedstock types and to iterate the models. Generally, ANN is superior to HLR in predicting the HHV of torrefied biomass with the optimized model achieving an R2 of 0.9133. Additionally, two biomasses subjected to two torrefaction technologies were used to validate the model, and an R2 of 0.9676 was achieved.As a result, in a torrefaction process, once the raw biomass is analyzed, the HHV of any torrefied biomass can be predicted simply by measuring the weight of the product.

3D Visual Grounding-Audio: 3D scene object detection based on audio

3D Visual Grounding (3DVG) is a prevalent multi-modal information fusion task capable of accurately localizing target objects referenced in natural language descriptions within a point cloud scene. Nevertheless, the stringent demands for input and output devices present substantial hurdles for the application and integration of 3DVG in fields like remote robotic control and telemedicine. To address this challenge, we introduce several innovative approaches. Firstly, we have initiated a novel multi-modal task, termed 3D Visual Grounding-Audio (3DVG-Audio), which is based on the fusion of audio and point cloud. To the best of our knowledge, this represents the first instance of an Audio-Point Cloud multi-modal task. 3DVG-Audio achieves precise localization of audio-mentioned objects within the point cloud by utilizing the point cloud in conjunction with the corresponding audio input. Secondly, building upon the ScanRefer, we have developed a novel dataset named 3DVG-AudioSet, specifically designed for the training and evaluation of the 3DVG-Audio method. Finally, we have crafted a tailored loss function to further enhance the performance of 3DVG-Audio and introduced a method named AP-Refer, which serves as a benchmark for the task. Extensive experimental results demonstrate the potential for deep integration of audio and point cloud to tackle complex real-world challenges. AP-Refer has successfully addressed the 3DVG-Audio, circumventing the limitations of conventional 3DVG methods, and exhibits significant application potential.

Global–Local Query-Support Cross-Attention for Few-Shot Semantic Segmentation

Few-shot semantic segmentation (FSS) models aim to segment unseen target objects in a query image with scarce annotated support samples. This challenging task requires an effective utilization of support information contained in the limited support set. However, the majority of existing FSS methods either compressed support features into several prototype vectors or constructed pixel-wise support-query correlations to guide the segmentation, which failed in effectively utilizing the support information from the global–local perspective. In this paper, we propose Global–Local Query-Support Cross-Attention (GLQSCA), where both global semantics and local details are exploited. Implemented with multi-head attention in a transformer architecture, GLQSCA treats every query pixel as a token, aggregates the segmentation label from the support mask values (weighted by the similarities with all foreground prototypes (global information)), and supports pixels (local information). Experiments show that our GLQSCA significantly surpasses state-of-the-art methods on the standard FSS benchmarks PASCAL-5i and COCO-20i.

Stabilizing Dynamic Backscatter for Swift and Accurate Object Tracking

Accurate and high-speed object movement tracking systems often face significant challenges due to signal instability caused by the object’s movement and rotation. To address these issues, we present STABack, a novel object tracking system using backscatter tags and accelerometer sensors, designed for high-accuracy, high-speed movement tracking. We developed an amplitude stabilization algorithm which uses envelope detection to reduce the impact of high-frequency movement on the signal, and uses dynamic threshold output to lower the BER in backscatter demodulation. Our method reduces the BER by 0.3757 compared to the regular demodulation method, resulting in a final BER of 0.07. Our evaluation of the STABack prototype shows that it achieves a median distance measurement accuracy of 6.45 cm with a standard deviation of 6.95 cm, under the condition of a speed of 120 cm. The attitude angle estimation’s mean error is under 7 degrees. The system’s accuracy in detecting the target object’s trajectory is as high as 99%, and it can still decode with a bit error rate of no more than 0.034 at a speed of 166 cm/s. The power consumption of our system prototype is only 38.54 μW based on our experimental results. Overall, our results demonstrate that STABack can accurately estimate the movement and rotation of target objects in unstable backscatter channels.

Individual differences in proprioceptive reorientation: a study on body characteristics and Posturography

ABSTRACT Spatial reorientation guided by the proprioceptive information derived from walking on a slanted floor has revealed large individual and sex differences. Here we addressed the role of body characteristics and measures of stability and balance-related sensory function from computerized dynamic posturography. Blindfolded participants walked in a slanted arena and had to identify the uphill direction (Perceptual task) and use the slope to remember the location of a target object (Memory task). Male participants performed better in the Perceptual task. In the Memory task, heavier participants made smaller errors, suggesting that body mass may influence how we navigate in the environment.

Deep Learning-Based Biomimetic Identification Method for Mask Wearing Standardization.

Deep learning technology can automatically learn features from large amounts of data, with powerful feature extraction and pattern recognition capabilities, thereby improving the accuracy and efficiency of object detection. [The objective of this study]: In order to improve the accuracy and speed of mask wearing deep learning detection models in the post pandemic era, the [Problem this study aimed to resolve] was based on the fact that no research work has been reported on standardized detection models for mask wearing with detecting nose targets specially. [The topic and method of this study]: A mask wearing normalization detection model (towards the wearing style exposing the nose to outside, which is the most obvious characteristic of non-normalized style) based on improved YOLOv5s (You Only Look Once v5s is an object detection network model) was proposed. [The improved method of the proposed model]: The improvement design work of the detection model mainly includes (1) the BottleneckCSP (abbreviation of Bottleneck Cross Stage Partial) module was improved to a BottleneckCSP-MASK (abbreviation of Bottleneck Cross Stage Partial-MASK) module, which was utilized to replace the BottleneckCSP module in the backbone architecture of the original YOLOv5s model, which reduced the weight parameters' number of the YOLOv5s model while ensuring the feature extraction effect of the bonding fusion module. (2) An SE module was inserted into the proposed improved model, and the bonding fusion layer in the original YOLOv5s model was improved for better extraction of the features of mask and nose targets. [Results and validation]: The experimental results indicated that, towards different people and complex backgrounds, the proposed mask wearing normalization detection model can effectively detect whether people are wearing masks and whether they are wearing masks in a normalized manner. The overall detection accuracy was 99.3% and the average detection speed was 0.014 s/pic. Contrasted with original YOLOv5s, v5m, and v5l models, the detection results for two types of target objects on the test set indicated that the mAP of the improved model increased by 0.5%, 0.49%, and 0.52%, respectively, and the size of the proposed model compressed by 10% compared to original v5s model. The designed model can achieve precise identification for mask wearing behaviors of people, including not wearing a mask, normalized wearing, and wearing a mask non-normalized.

Can 5 Minutes of Finger Actions Boost Creative Incubation?

AbstractPrevious studies suggest that the activation of the motor system – either via action, motor imagery, or brain stimulation – may increase subsequent performance on divergent thinking tasks (e.g., the alternate uses task; AUT). We tested this idea in a within-subjects design by administering the AUT using four different target objects and four different 5-min incubation tasks that differed in terms of arm and finger movements. In between-subjects designs, 0-back incubation has been shown to yield more creative responses than rest. Additionally, we included two new incubations that both involved arm actions, but differed in the amount of finger actions (Iranian dance, ballet dance). Incubation tasks involving finger actions (Iranian dance, 0-back) were predicted to increase creativity for objects that are typically manipulated with the fingers. There was a main effect of object. Alternate uses given to the paperclip were rated as more creative than those given to the other objects. With our within-subjects design, we could not replicate the previously described difference between 0-back and rest incubations. However, hypothesis-driven comparisons showed that, although the interaction of object and incubation was not significant, Iranian dance yielded more creative usages for paperclip than for sheet of paper, cup and brick, and all other incubations yielded more creative usages for paperclip than for brick. Iranian dance also generated marginally more creative usages than ballet. Our results suggest that if the hypothesized effects exist, they are likely to be small. Overall, AUT performance seems more influenced by the AUT object than by type of incubation.

Zero-shot visual grounding via coarse-to-fine representation learning

Visual grounding (VG) locates target objects in visual scenes by understanding given natural language queries. Current methods for VG mainly focus on grounding referring expressions or noun phrases covered in the labeled training samples. Despite their grounding prowess, these approaches struggle in grounding novel query–image pairs excluded from the training data. This shortage is usually caused by the deficiency of discriminative representation learning both in images and queries. To address these issues, we propose a one-stage coarse-to-fine framework for zero-shot VG to ground novel query-image samples. Specifically, in the coarse stage, we mine the global context information in the visual features and query embeddings by employing a multi-head self-attention block, strengthening the intra-modality relations in the visual and textual features. In the fine stage, we first learn the query-aware visual representations based on the acquired global context information via a multi-modal relation-enhanced transformer block, which explores the informative information by modeling the cross-modal interaction among visual and textual domains. We further excavate target-oriented discriminative representations from the acquired query-aware visual representations by a noun phrase-guided multi-modal interaction network, which augments the interaction between target-related phrases and the obtained query-aware visual representations to enhance the distinction of target regions, enhancing the subsequent referred target grounding and generalizing. In order to validate the proposed approach, we implement extensive experiments and ablation studies on public benchmark datasets, including RefCOCO, RefCOCO+, RefCOCOg, Flickr30K Entity, Flickr-Split-0 and Flickr-Split-1. Experimental results demonstrate that our approach substantially improves the grounding accuracy and achieves new state-of-the-art performance under single-stage training and testing.

Research on a small target object detection method for aerial photography based on improved YOLOv7

Research on a small target object detection method for aerial photography based on improved YOLOv7

A Robust Tri-Electromagnet-Based 6-DoF Pose Tracking System Using an Error-State Kalman Filter.

Magnetic pose tracking is a non-contact, accurate, and occlusion-free method that has been increasingly employed to track intra-corporeal medical devices such as endoscopes in computer-assisted medical interventions. In magnetic pose-tracking systems, a nonlinear estimation algorithm is needed to recover the pose information from magnetic measurements. In existing pose estimation algorithms such as the extended Kalman filter (EKF), the 3-DoF orientation in the S3 manifold is normally parametrized as unit quaternions and simply treated as a vector in the Euclidean space, which causes a violation of the unity constraint of quaternions and reduces pose tracking accuracy. In this paper, a pose estimation algorithm based on the error-state Kalman filter (ESKF) is proposed to improve the accuracy and robustness of electromagnetic tracking systems. The proposed system consists of three electromagnetic coils for magnetic field generation and a tri-axial magnetic sensor attached to the target object for field measurement. A strategy of sequential coil excitation is developed to separate the magnetic fields from different coils and reject magnetic disturbances. Simulation and experiments are conducted to evaluate the pose tracking performance of the proposed ESKF algorithm, which is also compared with standard EKF and constrained EKF. It is shown that the ESKF can effectively maintain the quaternion unity and thus achieve a better tracking accuracy, i.e., a Euclidean position error of 2.23 mm and an average orientation angle error of 0.45°. The disturbance rejection performance of the electromagnetic tracking system is also experimentally validated.

Is there a cost when predictions are not met? A VWP study investigating L1 and L2 speakers.

Research has found that both first language (L1) and second language (L2) speakers make predictions about upcoming linguistic information, with predictive behaviour being impacted by individual differences and methodological factors. However, it is not clear whether a cost is incurred when a prediction is made, but not met. L2 speakers have less experience with their L2 and parsing can be cognitively demanding, which together may lead L2 speakers to incur prediction costs differently relative to L1 speakers. In this study using the visual world paradigm, we test whether L1 and L2 speakers predict in the same way, within the same time frame, and incur the same costs if predictions are not met. We also explore the role of proficiency and speech rate. We found that both groups predict in a similar way and within a similar time frame. In addition, neither group incurred a prediction cost when the target was the most likely alternative, though L2 speakers take longer to shift their attention to the target object when predictions are not met. We argue that this reflects a slowing of lexical access rather than a specific cost of prediction. We only found prediction differences when speech rate was included in the analysis, highlighting the importance of attending to speech rate in studies using the visual world paradigm. Overall, this study supports research showing that both L1 and L2 speakers may make multiple partial predictions about upcoming information rather than predicting one specific lexical candidate while inhibiting less likely lexical candidates.

VLAI: Exploration and Exploitation based on Visual-Language Aligned Information for Robotic Object Goal Navigation

Object Goal Navigation(ObjectNav) is the task that an agent need navigate to an instance of a specific category in an unseen environment through visual observations within limited time steps. This work plays a significant role in enhancing the efficiency of locating specific items in indoor spaces and assisting individuals in completing various tasks, as well as providing support for people with disabilities. To achieve efficient ObjectNav in unfamiliar environments, global perception capabilities, understanding the regularities of space and semantics in the environment layout are significant. In this work, we propose an explicit-prediction method called VLAI that utilizes visual-language alignment information to guide the agent's exploration, unlike previous navigation methods based on frontier potential prediction or egocentric map completion, which only leverage visual observations to construct semantic maps, thus failing to help the agent develop a better global perception. Specifically, when predicting long-term goals, we retrieve previously saved visual observations to obtain visual information around the frontiers based on their position on the incrementally built incomplete semantic map. Then, we apply our designed Chat Describer to this visual information to obtain detailed frontier object descriptions. The Chat Describer, a novel automatic-questioning approach deployed in Visual-to-Language, is composed of Large Language Model(LLM) and the visual-to-language model(VLM), which has visual question-answering functionality. In addition, we also obtain the semantic similarity of target object and frontier object categories. Ultimately, by combining the semantic similarity and the boundary descriptions, the agent can predict the long-term goals more accurately. Our experiments on the Gibson and HM3D datasets reveal that our VLAI approach yields significantly better results compared to earlier methods. The code is released athttps://github.com/31539lab/VLAI.

Depth range extended digital holography for precise 3D profile imaging via dual frequency interval sweeping

A 3D profile imaging method is proposed by using digital holography swept at dual frequency intervals. The dual frequency interval sweeping is achieved by altering the frequency of the tunable laser with two selected frequency intervals. An infrared camera was used to capture the interferometric patterns during two sweeping periods, and depth profile was reconstructed from multiple digital holograms. Phase image across the target region is extracted at each frequency and the phase at each pixel is varied as the frequencies are swept at a prefixed interval in each period. From measured phases, the congruence equations are constructed to achieve the 3D profile of the target object. The proposed sweeping method extends the measurement range by about ten times compared to the classical single frequency interval sweeping method, with a high depth accuracy of within 20 μm over a range of 4 cm. The proposed method is verified by both numerical simulation and experiment. Automatic extraction of reconstructed distances is achieved to facilitate automated 3D profile imaging.