Target Localization Task Research Articles

LAACNet: Lightweight adaptive activation convolution network-based defect detection on polished metal surfaces

After the metal workpiece has been polished, there may still be small defects on the surface, which will adversely affect the quality of the product and the service life and availability of the metal in severe cases. To solve this problem, the industrial sector has been seeking more lightweight and efficient solutions. Therefore, this thesis proposes a Lightweight Adaptive Activation Convolution Network (LAACNet). Firstly, this thesis proposes a lightweight convolution module that adopts features concatenation to realize intra-channel and inter-channel information transfer and fusion. Secondly, this thesis proposes an adaptive activation convolution module with the enhanced nonlinear expression of the module, which makes the deep neural network expression more powerful. This thesis proposes a spatial channel coordinate attention module to capture the long-range dependencies between image pixels better. Finally, this thesis introduces a loss function that can optimize performance in target classification and localization tasks. Experiments were conducted on the self-made datasets Metal Surface Defect-Detection (MSD-DET) and GC10-Detection (GC10-DET), achieving Mean Average Precision 50 (mAP50) of 86.3% and 66.8%, respectively. The detection performance of this model is superior to other methods. In the ablation experiment, this thesis verified the effectiveness of each module. This thesis validated the Northeastern University-Detection (NEU-DET) dataset, achieving a mAP50 of 76.0%. The results show that LAACNet exhibits excellent robustness and generalization performance in surface defect detection. In addition, the method significantly reduces the number of model parameters, providing an effective choice for lightweight and efficient solutions.

Object Localization Algorithm Based on Meta-Reinforcement Learning

Abstract When the target localization algorithm based on reinforcement learning is trained on few-sample data sets, the accuracy of target localization is low due to the low degree of fitting. Therefore, on the basis of deep reinforcement learning target localization algorithm, this paper proposes a target localization algorithm based on meta-reinforcement learning. Firstly, during the initial training of the model, the meta-parameters were classified and stored according to the similarity of the training tasks. Then, for the new target location task, the task feature extraction was carried out and the meta parameters with the highest similarity were matched as the initial parameters of the model training. The model dynamically updated the meta parameter pool to ensure that the optimal meta parameters of multiple different types of features were saved in the meta parameter pool, so as to improve the generalization ability and recognition accuracy of multiple types of target location tasks. Experimental results show that in a variety of single target localization tasks, compared with the original reinforcement learning target localization algorithm, under the same data set size, the model converges under a small number of training steps with the meta-parameters in the matching meta-parameter pool as the initial training parameters. Moreover, the training speed of the meta-reinforcement learning method based on MAML-RL is increased by 28.2% for random initial parameters, and that of the meta-reinforcement learning method based on this paper is increased by 34.9%, indicating that the proposed algorithm effectively improves the training speed, generalization performance and localization accuracy of object detection.

Automatic weld type classification, tacked spot recognition and weld ROI determination for robotic welding based on modified YOLOv5

Automatic and fast weld type classification, tacked spot recognition and weld ROI (region of interest) determination are key links for intelligent welding robot, because they directly affect the seam tracking and welding parameters such as current, voltage and torch inclination. Nevertheless, there are few studies on weld type classification, tacked spot recognition, and weld ROI determination. Considering the fast inference speed of YOLOv5, an automatic weld type classification, tacked spot recognition and weld ROI determination based on modified YOLOv5 is presented in this paper. First, the detection requirements of weld type classification, tacked spot recognition and weld ROI determination are transformed into a unified target localization task to improve the inference speed, so the three results can be obtained through single inference; the next, to improve the localization accuracy of weld ROI, the center component bias between the predicted box and ground truth is added to the original CIOU localization loss function; then, a weighted classification loss function is used to reduce the false positives in fillet and groove welds; finally, a self-template method for padding image border is presented to improve the generalization ability of the trained model. Experimental results show that: the presented method reaches 100% precision, 100% recall, 0.91 mean intersection-over-unio, 2.41 pixels center component bias of determined weld ROI and 18 ms inference times in the original size images; the center component bias of determined weld ROI is reduced from 2.38 pixels to 2.18 pixels by adding the center component bias loss to CIOU function in the padded images; the weighted classification loss function reduced the false positives in fillet and groove welds; compared with the default gray border padding method, the model trained by using self-template padding method reduced the center component bias of the determined weld ROI from 2.73 pixels to 2.41 pixels in the original size images. Moreover, when Ref. Chen et al. (2022) positions the welding seam coordinates in the weld ROI determined by the presented method, the recall is improved from 0.96 to 0.98, and the computation time is reduced from 180 ms to 48 ms.

A Moving Target Online Location Algorithm Based on Factor Graph

The superior performance of factor graphs compared to Kalman filtering in various fields and the use of factor graph algorithms instead of Kalman filtering algorithms in moving target localization tasks can reduce target localization error by more than 50%. However, the global factor graph algorithm may cause computational delays due to excessive computational effort. A moving target localization algorithm based on a combination of global and incremental optimization with improved factor graphs is proposed to improve localization accuracy and ensure that the computation time can be adapted to the requirements of online location. A reference point is introduced into the incremental calculation process, and it is first determined whether global or incremental calculation is used for this calculation by comparing the distance between the incremental localization results of the calculated reference point. The position of the UAV itself is then corrected by determining the position of the reference point, and this is used to finally locate the target. Simulation results show that the algorithm has good real-time performance compared to the time-consuming global algorithm. The online localization error of moving targets can be reduced by 17% compared to the incremental calculation results of the classical factor graph algorithm.

Trajectory Optimization of Multi-UAVs for Marine Target Tracking during Approaching Stage

This paper focuses on a coordinated tracking planning method of multiple unmanned aerial vehicles (UAVs), which was deployed in the best positions to better fulfill the marine target localization tasks when approaching the target. The optimal planning of multi-UAVs was implemented using an online centralized nonlinear model predictive control (NMPC) based on the target state’s uncertainty criteria. The penalty function is used to solve UAV platform dynamic performance in the model predictive control method to consider the more realistic situation. The coordinated planning problems of multi-UAVs are numerically simulated and compared with the Lyapunov vector field guidance (LVFG) method under classical mission scenarios. Simulation results demonstrate that the algorithm can maintain the optimal observation configuration of multi-UAVs to improve the marine target positioning accuracy, verifying the feasibility and superiority of this method. Furthermore, the simulation results can provide a useful reference for the flight control law design of multi-UAVs with optimal observation configuration.

The Effects of Rewards on Autonomous Unmanned Aerial Vehicle (UAV) Operations Using Reinforcement Learning

The effects of rewards on the ability of an autonomous UAV controlled by a Reinforcement Learning agent to accomplish a target localization task were investigated. It was shown that with an increase in the reward obtained by a learning agent upon correct detection, systems would become more risk-tolerant, efficient and have a tendency to locate targets faster with an increase in the sensor sensitivity after systems achieve steady-state performance.

The Gaze Cueing Effect and Its Enhancement by Facial Expressions Are Impacted by Task Demands: Direct Comparison of Target Localization and Discrimination Tasks.

The gaze cueing effect is characterized by faster attentional orienting to a gazed-at than a non-gazed-at target. This effect is often enhanced when the gazing face bears an emotional expression, though this finding is modulated by a number of factors. Here, we tested whether the type of task performed might be one such modulating factor. Target localization and target discrimination tasks are the two most commonly used gaze cueing tasks, and they arguably differ in cognitive resources, which could impact how emotional expression and gaze cues are integrated to orient attention. In a within-subjects design, participants performed both target localization and discrimination gaze cueing tasks with neutral, happy, and fearful faces. The gaze cueing effect for neutral faces was greatly reduced in the discrimination task relative to the localization task, and the emotional enhancement of the gaze cueing effect was only present in the localization task and only when this task was performed first. These results suggest that cognitive resources are needed for gaze cueing and for the integration of emotional expressions and gaze cues. We propose that a shift toward local processing may be the mechanism by which the discrimination task interferes with the emotional modulation of gaze cueing. The results support the idea that gaze cueing can be greatly modulated by top-down influences and cognitive resources and thus taps into endogenous attention. Results are discussed within the context of the recently proposed EyeTune model of social attention.

Effect of Sensor Sensitivity on Autonomous Aerial Vehicle Target Localization

In recent years, Unmanned Aerial Systems (UASs) development and application have achieved remarkable growth. With the advancement of technology, UASs nowadays feature more advanced autonomous capabilities than ever before. In order to achieve autonomous behavior, intelligent systems are required to be incorporated to support system learning, control and decision-making. With these capabilities, UASs can learn from their past experiences, through interacting with the task environment to adapt their behavior to enhance their future performance. Machine learning is one of the most commonly used techniques for UASs to acquire knowledge from their experience, and research in this area is still developing. In this study, Reinforcement Learning (RL) algorithms were used on autonomous aerial systems to achieve adaptive behavior and decision-making capabilities. The effects of UAS sensor sensitivity, as modeled through Signal Detection Theory (SDT), on the ability of RL algorithms to accomplish a target localization task were investigated. Three levels of sensor sensitivity were simulated and compared to the results of the same system using a perfect sensor, with the consideration of two RL algorithms, namely, Temporal Difference (TD) and Monte Carlo (MC) methods. Target localization and identification task were used as the test bed, and a hierarchical architecture was developed with two distinct agents. Mission performance was analyzed using multiple metrics, including episodic reward and the time taken to locate all targets. Statistical analyses were carried out to detect significant differences in the comparison of steady-state behavior of different factors. Results were discussed, and future research direction was given at the end of the paper.

Improving Multimodal Localization Through Self-Supervision

Modern warehouses utilize fleets of robots for inventory management. To ensure efficient and safe operation, real-time localization of each agent is essential. Most robots follow metal tracks buried in the floor and use a grid of precisely mounted RFID tags for localization. As robotic agents in warehouses and manufacturing plants become ubiquitous, it would be advantageous to eliminate the need for these metal wires and RFID tags. Not only do they suffer from significant installation costs, the removal of wires would allow agents to travel to any area inside the building. Sensors including cameras and LiDAR have provided meaningful localization information for many different positioning system implementations. Fusing localization features from multiple sensor sources is a challenging task especially when the target localization task’s dataset is small. We propose a deep-learning based localization system which fuses features from an omnidirectional camera image and a 3D LiDAR point cloud to create a robust robot positioning model. Although the usage of vision and LiDAR eliminate the need for the precisely installed RFID tags, they do require the collection and annotation of ground truth training data. Deep neural networks thrive on lots of supervised data, and the collection of this data can be time consuming. Using a dataset collected in a warehouse environment, we evaluate the performance of two individual sensor models for localization accuracy. To minimize the need for extensive ground truth data collection, we introduce a self-supervised pretraining regimen to populate the image feature extraction network with meaningful weights before training on the target localization task with limited data. In this research, we demonstrate how our self-supervision improves accuracy and convergence of localization models without the need for additional sample annotation.

Energy-efficient scheme for target recognition and localization in wireless acoustic sensor networks

The development of wireless acoustic sensor networks has driven the use of acoustic signals for target monitoring. Most monitoring applications require continuous network connectivity and data transfers, which can rapidly exhaust nodes’ energy. Consequently, sensors must collaborate in an adequate architecture to perform target recognition and localization tasks and then to send the results to a remote server with a reduced data volume. The design of an energy-efficient scheme that achieves acoustic target recognition and localization remains an open research problem. Accordingly, this article proposes a low-energy acoustic-based sensing scheme for target recognition and localization to be implemented in a cluster-based sensing approach designed to appropriately balance energy consumption and local processing performed by sensor nodes. A reduced set of low-complexity feature extraction methods in the time domain signal are used in the recognition process. The scheme uses the received energy of the acoustic signals for the target localization. This article details the network architecture, the scheme specification, and its implementation. The results show that the scheme can classify targets with 81.34% accuracy. It requires 3.2 mJ of energy when executed in MICAz, achieving 99% energy savings compared to streaming 3 s of an acoustic signal to a remote server.

Engineering Artificial Somatosensation Through Cortical Stimulation in Humans.

Sensory feedback is a critical aspect of motor control rehabilitation following paralysis or amputation. Current human studies have demonstrated the ability to deliver some of this sensory information via brain-machine interfaces, although further testing is needed to understand the stimulation parameters effect on sensation. Here, we report a systematic evaluation of somatosensory restoration in humans, using cortical stimulation with subdural mini-electrocorticography (mini-ECoG) grids. Nine epilepsy patients undergoing implantation of cortical electrodes for seizure localization were also implanted with a subdural 64-channel mini-ECoG grid over the hand area of the primary somatosensory cortex (S1). We mapped the somatotopic location and size of receptive fields evoked by stimulation of individual channels of the mini-ECoG grid. We determined the effects on perception by varying stimulus parameters of pulse width, current amplitude, and frequency. Finally, a target localization task was used to demonstrate the use of artificial sensation in a behavioral task. We found a replicable somatotopic representation of the hand on the mini-ECoG grid across most subjects during electrical stimulation. The stimulus-evoked sensations were usually of artificial quality, but in some cases were more natural and of a cutaneous or proprioceptive nature. Increases in pulse width, current strength and frequency generally produced similar quality sensations at the same somatotopic location, but with a perception of increased intensity. The subjects produced near perfect performance when using the evoked sensory information in target acquisition tasks. These findings indicate that electrical stimulation of somatosensory cortex through mini-ECoG grids has considerable potential for restoring useful sensation to patients with paralysis and amputation.

Affective blindsight in the absence of input from face processing regions in occipital-temporal cortex

Previous research suggests that the implicit recognition of emotional expressions may be carried out by pathways that bypass primary visual cortex (V1) and project to the amygdala. Some of the strongest evidence supporting this claim comes from case studies of “affective blindsight” in which patients with V1 damage can correctly guess whether an unseen face was depicting a fearful or happy expression. In the current study, we report a new case of affective blindsight in patient MC who is cortically blind following extensive bilateral lesions to V1, as well as face and object processing regions in her ventral visual stream. Despite her large lesions, MC has preserved motion perception which is related to sparing of the motion sensitive region MT+ in both hemispheres.To examine affective blindsight in MC we asked her to perform gender and emotion discrimination tasks in which she had to guess, using a two-alternative forced-choice procedure, whether the face presented was male or female, happy or fearful, or happy or angry. In addition, we also tested MC in a four-alternative forced-choice target localization task. Results indicated that MC was not able to determine the gender of the faces (53% accuracy), or localize targets in a forced-choice task. However, she was able to determine, at above chance levels, whether the face presented was depicting a happy or fearful (67%, p = .006), or a happy or angry (64%, p = .025) expression. Interestingly, although MC was better than chance at discriminating between emotions in faces when asked to make rapid judgments, her performance fell to chance when she was asked to provide subjective confidence ratings about her performance. These data lend further support to the idea that there is a non-conscious visual pathway that bypasses V1 which is capable of processing affective signals from facial expressions without input from higher-order face and object processing regions in the ventral visual stream.

Temporal characteristics of contextual effects in sound localization.

Two experiments examined plasticity induced by context in a simple target localization task. The context was represented by interleaved localization trials with the target preceded by a distractor. In a previous study, the context induced large response shifts when the target and distractor stimuli were identical 2-ms-noise clicks [Kopčo, Best, and Shinn-Cunningham (2007). J. Acoust. Soc. Am. 121, 420-432]. Here, the temporal characteristics of the contextual effect were examined for the same stimuli. Experiment 1 manipulated the context presentation rate and the distractor-target inter-stimulus interval (ISI). Experiment 2 manipulated the temporal structure of the context stimulus, replacing the one-click distractor either by a distractor consisting of eight sequentially presented clicks or by a noise burst with total energy and duration identical to the eight-click distractor. In experiment 1, the contextual shift size increased with increasing context rate while being largely independent of ISI. In experiment 2, the eight-click-distractor induced a stronger shift than the one-click-distractor context, while the noise-distractor context induced a very small shift. These results suggest that contextual plasticity is an adaptation driven both by low-level factors like spatiotemporal context distribution and higher-level factors like perceptual similarity between the stimuli, possibly related to precedence buildup.

Adaptive Environmental Source Localization and Tracking with Unknown Permittivity and Path Loss Coefficients.

Accurate signal-source and signal-reflector target localization tasks via mobile sensory units and wireless sensor networks (WSNs), including those for environmental monitoring via sensory UAVs, require precise knowledge of specific signal propagation properties of the environment, which are permittivity and path loss coefficients for the electromagnetic signal case. Thus, accurate estimation of these coefficients has significant importance for the accuracy of location estimates. In this paper, we propose a geometric cooperative technique to instantaneously estimate such coefficients, with details provided for received signal strength (RSS) and time-of-flight (TOF)-based range sensors. The proposed technique is integrated to a recursive least squares (RLS)-based adaptive localization scheme and an adaptive motion control law, to construct adaptive target localization and adaptive target tracking algorithms, respectively, that are robust to uncertainties in aforementioned environmental signal propagation coefficients. The efficiency of the proposed adaptive localization and tracking techniques are both mathematically analysed and verified via simulation experiments.

Nonuniform Changes in the Distribution of Visual Attention from Visual Complexity and Action: A Driving Simulation Study.

Researchers acknowledge the interplay between action and attention, but typically consider action as a response to successful attentional selection or the correlation of performance on separate action and attention tasks. We investigated how concurrent action with spatial monitoring affects the distribution of attention across the visual field. We embedded a functional field of view (FFOV) paradigm with concurrent central object recognition and peripheral target localization tasks in a simulated driving environment. Peripheral targets varied across 20-60 deg eccentricity at 11 radial spokes. Three conditions assessed the effects of visual complexity and concurrent action on the size and shape of the FFOV: (1) with no background, (2) with driving background, and (3) with driving background and vehicle steering. The addition of visual complexity slowed task performance and reduced the FFOV size but did not change the baseline shape. In contrast, the addition of steering produced not only shrinkage of the FFOV, but also changes in the FFOV shape. Nonuniform performance decrements occurred in proximal regions used for the central task and for steering, independent of interference from context elements. Multifocal attention models should consider the role of action and account for nonhomogeneities in the distribution of attention.

Hemispheric asymmetry in the auditory facilitation effect in dual-stream rapid serial visual presentation tasks.

Even though auditory stimuli do not directly convey information related to visual stimuli, they often improve visual detection and identification performance. Auditory stimuli often alter visual perception depending on the reliability of the sensory input, with visual and auditory information reciprocally compensating for ambiguity in the other sensory domain. Perceptual processing is characterized by hemispheric asymmetry. While the left hemisphere is more involved in linguistic processing, the right hemisphere dominates spatial processing. In this context, we hypothesized that an auditory facilitation effect in the right visual field for the target identification task, and a similar effect would be observed in the left visual field for the target localization task. In the present study, we conducted target identification and localization tasks using a dual-stream rapid serial visual presentation. When two targets are embedded in a rapid serial visual presentation stream, the target detection or discrimination performance for the second target is generally lower than for the first target; this deficit is well known as attentional blink. Our results indicate that auditory stimuli improved target identification performance for the second target within the stream when visual stimuli were presented in the right, but not the left visual field. In contrast, auditory stimuli improved second target localization performance when visual stimuli were presented in the left visual field. An auditory facilitation effect was observed in perceptual processing, depending on the hemispheric specialization. Our results demonstrate a dissociation between the lateral visual hemifield in which a stimulus is projected and the kind of visual judgment that may benefit from the presentation of an auditory cue.

Visuospatial selective attention in chickens.

Voluntary control of attention promotes intelligent, adaptive behaviors by enabling the selective processing of information that is most relevant for making decisions. Despite extensive research on attention in primates, the capacity for selective attention in nonprimate species has never been quantified. Here we demonstrate selective attention in chickens by applying protocols that have been used to characterize visual spatial attention in primates. Chickens were trained to localize and report the vertical position of a target in the presence of task-relevant distracters. A spatial cue, the location of which varied across individual trials, indicated the horizontal, but not vertical, position of the upcoming target. Spatial cueing improved localization performance: accuracy (d') increased and reaction times decreased in a space-specific manner. Distracters severely impaired perceptual performance, and this impairment was greatly reduced by spatial cueing. Signal detection analysis with an "indecision" model demonstrated that spatial cueing significantly increased choice certainty in localizing targets. By contrast, error-aversion certainty (certainty of not making an error) remained essentially constant across cueing protocols, target contrasts, and individuals. The results show that chickens shift spatial attention rapidly and dynamically, following principles of stimulus selection that closely parallel those documented in primates. The findings suggest that the mechanisms that control attention have been conserved through evolution, and establish chickens--a highly visual species that is easily trained and amenable to cutting-edge experimental technologies--as an attractive model for linking behavior to neural mechanisms of selective attention.

An attentional-adaptation account of spatial negative priming: evidence from event-related potentials.

Negative priming (NP) refers to a slower response to a target stimulus if it has been previously ignored. To examine theoretical accounts of spatial NP, we recorded behavioral measures and event-related potentials (ERPs) in a target localization task. A target and distractor briefly appeared, and the participant pressed a key corresponding to the target's location. The probability of the distractor appearing in each of four locations varied, whereas the target appeared with equal probabilities in all locations. We found that response times (RTs) were fastest when the prime distractor appeared in its most probable (frequent) location and when the prime target appeared in the location that never contained a distractor. Moreover, NP effects varied as a function of location: They were smallest when targets followed distractors in the frequent distractor location-a finding not predicted by episodic-retrieval or suppression accounts of NP. The ERP results showed that the P2, an ERP component associated with attentional orientation, was smaller in prime displays when the distractor appeared in its frequent location. Moreover, no differences were apparent between negative-prime and control trials in the N2, which is associated with suppression processes, nor in the P3, which is associated with episodic retrieval processes. These results indicate that the spatial NP effect is caused by both short- and long-term adaptation in preferences based on the history of inspecting unsuccessful locations. This article is dedicated to the memory of Edward E. Smith, and we indicate how this study was inspired by his research career.

Perception of emergent configurations in humans (Homo sapiens) and chimpanzees (Pan troglodytes).

We examined the perceptions of emergent configurations in humans and chimpanzees using a target-localization task. The stimulus display consisted of a target placed among multiple identical distractors. The target and distractors were presented either solely, within congruent contexts in which salient configurations emerge, or within incongruent contexts in which salient configurations do not emerge. We found that congruent contexts had similar facilitative effects on target localization by humans and chimpanzees, whereas similar disruptive effects emerged when the stimuli were presented within incongruent contexts. When display size was manipulated, targets under the congruent-context condition were localized in a parallel manner, but those under the no-context and incongruent-context conditions were localized in a serial manner by both species. These results suggest that both humans and chimpanzees perceive emergent configurations when targets and distractors are presented within certain congruent contexts and that they process such emergent configurations preattentively.

Automatic Selective Attention as a Function of Sensory Modality in Aging

It was recently hypothesized that age-related differences in selective attention depend on sensory modality (Guerreiro, M. J. S., Murphy, D. R., & Van Gerven, P. W. M. (2010). The role of sensory modality in age-related distraction: A critical review and a renewed view. Psychological Bulletin, 136, 975-1022. doi:10.1037/a0020731). So far, this hypothesis has not been tested in automatic selective attention. The current study addressed this issue by investigating age-related differences in automatic spatial cueing effects (i.e., facilitation and inhibition of return [IOR]) across sensory modalities. Thirty younger (mean age = 22.4 years) and 25 older adults (mean age = 68.8 years) performed 4 left-right target localization tasks, involving all combinations of visual and auditory cues and targets. We used stimulus onset asynchronies (SOAs) of 100, 500, 1,000, and 1,500 ms between cue and target. The results showed facilitation (shorter reaction times with valid relative to invalid cues at shorter SOAs) in the unimodal auditory and in both cross-modal tasks but not in the unimodal visual task. In contrast, there was IOR (longer reaction times with valid relative to invalid cues at longer SOAs) in both unimodal tasks but not in either of the cross-modal tasks. Most important, these spatial cueing effects were independent of age. The results suggest that the modality hypothesis of age-related differences in selective attention does not extend into the realm of automatic selective attention.