Search Results (14)

Figure 1
<p>The UAV motion model.</p> Full article ">Figure 2
<p>Training process of PPO.</p> Full article ">Figure 3
<p>Network structure diagram.</p> Full article ">Figure 4
<p>Framework of H-PPO.</p> Full article ">Figure 5
<p>Diagram of UAV detection areas.</p> Full article ">Figure 6
<p>UAV detection model.</p> Full article ">Figure 7
<p>A schematic diagram of the experimental task area.</p> Full article ">Figure 8
<p>Reward function curves for different learning rates.</p> Full article ">Figure 9
<p>Maneuver reward curve in training phase (5 obstacles).</p> Full article ">Figure 10
<p>Maneuver reward curve in training phase (10 obstacles).</p> Full article ">Figure 11
<p>Maneuver reward curve in training phase (15 obstacles).</p> Full article ">Figure 12
<p>Number of movable threat targets and success rate.</p> Full article ">Figure 13
<p>Intelligent agent maneuver test diagram.</p> Full article ">Figure 14
<p>Dynamic scene test diagram.</p> Full article ">Figure 15
<p>Relationship between success rate and maximum maneuvering speed.</p> Full article ">

Figure 1
<p>Scenario of UAV swarm-based edge computing.</p> Full article ">Figure 2
<p>3D coordinate-coupled dynamic model of the UAV swarm’s centroid.</p> Full article ">Figure 3
<p>Network structure for maneuver model estimation.</p> Full article ">Figure 4
<p>Simulated UAV swarm trajectories for network training.</p> Full article ">Figure 5
<p>The tracking scenario used for validating the centroid trajectory tracking.</p> Full article ">Figure 6
<p>Training process of the model estimation network.</p> Full article ">Figure 7
<p>True mode and estimated mode probabilities of the maneuvering centroid. The topmost subgraph represents the true mode. The subgraphs below compare IMM estimates and GRU estimates for each model.</p> Full article ">Figure 8
<p>RMSEs of the position estimates for Monte Carlo simulation of the proposed method and conventional method.</p> Full article ">Figure 9
<p>Comparison of different centroid tracking methods. The subgraph at the top left is a view of the entire flight path. The remaining three subgraphs are fragments selected from the track, and the corresponding areas are shown in the box in the figure.</p> Full article ">Figure 10
<p>(<b>a</b>) Picture of the UAV. (<b>b</b>) Picture of the cooperative flight.</p> Full article ">Figure 11
<p>Flight trajectories with different durations of (<b>a</b>) 1194 s, (<b>b</b>) 1243 s, and (<b>c</b>) 1223 s.</p> Full article ">Figure 12
<p>Confusion matrix of three flight experiments with networks trained by simulation and actual observations. (<b>a</b>) The 1194 s flight with a simulated network. (<b>b</b>) The 1243 s flight with a simulated network. (<b>c</b>) The 1223 s flight with a simulated network. (<b>d</b>) The 1194 s flight with a retrained network. (<b>e</b>) The 1243 s flight with a retrained network. (<b>f</b>) The 1223 s flight with a retrained network.</p> Full article ">

Figure 1
<p>Scheme of the roll–pitch seeker.</p> Full article ">Figure 2
<p>Diagram of the seeker detector error angle.</p> Full article ">Figure 3
<p>Coordinate transformation relationship.</p> Full article ">Figure 4
<p>Stable loop of the roll frame.</p> Full article ">Figure 5
<p>Stable loop of the pitch frame.</p> Full article ">Figure 6
<p>Projection of the disturbance on the pitch frame coordinate.</p> Full article ">Figure 7
<p>Principle of the existence of the angular velocity of optical axis disturbance.</p> Full article ">Figure 8
<p>Schematic diagram of detector measurement error angle.</p> Full article ">Figure 9
<p>Schematic diagram of image changes on the detector: (<b>a</b>) Roll frame motion; (<b>b</b>) Pitch frame motion.</p> Full article ">Figure 10
<p>Principle of roll–pitch seeker stable control system.</p> Full article ">Figure 11
<p>Schematic diagram of coordinate transformation of roll–pitch seeker.</p> Full article ">Figure 12
<p>Frame control model.</p> Full article ">Figure 13
<p>DRR generated by roll interference torque: (<b>a</b>) damping torque; (<b>b</b>) spring torque.</p> Full article ">Figure 14
<p>Bode diagram of DRR function under different interference forces: (<b>a</b>) damping torque; (<b>b</b>) spring torque.</p> Full article ">Figure 15
<p>Roll frame angle.</p> Full article ">Figure 16
<p>Pitch frame angle.</p> Full article ">Figure 17
<p>Angular rate of roll frame.</p> Full article ">Figure 18
<p>Angular rate of pitch frame.</p> Full article ">Figure 19
<p>Optical axis pitch LOS rate.</p> Full article ">Figure 20
<p>Optical axis yaw LOS rate.</p> Full article ">Figure 21
<p>Error angle: (<b>a</b>) Detector error angle; (<b>b</b>) Pitch and yaw error angle of the Detector.</p> Full article ">Figure 22
<p>Roll frame angle.</p> Full article ">Figure 23
<p>Pitch frame angle.</p> Full article ">Figure 24
<p>Angular rate of roll frame.</p> Full article ">Figure 25
<p>Angular rate of pitch frame.</p> Full article ">Figure 26
<p>Optical axis pitch LOS rate.</p> Full article ">Figure 27
<p>Optical axis yaw LOS rate.</p> Full article ">Figure 28
<p>Detector error angle.</p> Full article ">Figure 29
<p>Pitch frame Monte Carlo simulation.</p> Full article ">Figure 30
<p>Optical axis pitch LOS rate.</p> Full article ">Figure 31
<p>Optical axis yaw LOS rate.</p> Full article ">Figure 32
<p>Error angle: (<b>a</b>) Detector error angle; (<b>b</b>) Pitch and yaw error angle of the Detector.</p> Full article ">Figure 33
<p>Pitch frame angle.</p> Full article ">Figure 34
<p>Angular rate of pitch frame.</p> Full article ">Figure 35
<p>Optical axis pitch LOS rate.</p> Full article ">Figure 36
<p>Detector error angle.</p> Full article ">Figure 37
<p>Pitch frame angle.</p> Full article ">Figure 38
<p>Angular rate of pitch frame.</p> Full article ">Figure 39
<p>Optical axis pitch LOS rate.</p> Full article ">Figure 40
<p>Detector error angle.</p> Full article ">Figure 41
<p>Pitch frame Monte Carlo simulation.</p> Full article ">Figure 42
<p>Roll frame angle.</p> Full article ">Figure 43
<p>Pitch frame angle.</p> Full article ">Figure 44
<p>Angular rate of roll frame.</p> Full article ">Figure 45
<p>Angular rate of pitch frame.</p> Full article ">Figure 46
<p>Optical axis pitch LOS rate.</p> Full article ">Figure 47
<p>Optical axis yaw LOS rate.</p> Full article ">Figure 48
<p>Error angle: (<b>a</b>) Detector error angle; (<b>b</b>) Pitch and yaw error angle of the Detector.</p> Full article ">Figure 49
<p>Monte Carlo simulation: (<b>a</b>) roll frame; (<b>b</b>) pitch frame.</p> Full article ">Figure 50
<p>Roll frame angle.</p> Full article ">Figure 51
<p>Pitch frame angle.</p> Full article ">Figure 52
<p>Angular rate of roll frame.</p> Full article ">Figure 53
<p>Angular rate of pitch frame.</p> Full article ">Figure 54
<p>Optical axis pitch LOS rate.</p> Full article ">Figure 55
<p>Optical axis yaw LOS rate.</p> Full article ">Figure 56
<p>Error angle: (<b>a</b>) Detector error angle; (<b>b</b>) Pitch and yaw error angle of the Detector.</p> Full article ">Figure 57
<p>Detector error angle in case the DRR exists in the pitch frame.</p> Full article ">Figure 58
<p>Detector error angle in case the DRR exists in the roll frame.</p> Full article ">Figure 59
<p>Pitch frame angle.</p> Full article ">Figure 60
<p>Angular rate of pitch frame.</p> Full article ">Figure 61
<p>Optical axis pitch LOS rate.</p> Full article ">Figure 62
<p>Optical axis yaw LOS rate.</p> Full article ">Figure 63
<p>Error angle: (<b>a</b>) Detector error angle; (<b>b</b>) Pitch and yaw error angle of the Detector.</p> Full article ">Figure 64
<p>Monte Carlo simulation: (<b>a</b>) roll frame; (<b>b</b>) pitch frame.</p> Full article ">

Figure 1
<p>Post-cardiac arrest management.</p> Full article ">Figure 2
<p>Neuroprognostication algorithm. Abbreviations: EEG—electroencephalogram, SSEP—somatosensory evoked potentials, NSE—neuron-specific enolase, CT—computerized tomography, MRI—Magnetic resonance Imaging. Legends: (1) suppressed background with or without periodic discharges and burst-suppression; (2) Bilateral absence of somatosensory evoked cortical N20-potentials.</p> Full article ">

Figure 1
<p>The schematic diagram of AOU for moving target.</p> Full article ">Figure 2
<p>Target and observer position with estimates.</p> Full article ">Figure 3
<p>Comparison of estimated velocity.</p> Full article ">Figure 4
<p>Comparison of estimated course.</p> Full article ">Figure 5
<p>Comparison of RMSE for estimated velocity and course.</p> Full article ">Figure 6
<p>Comparison of RMSE for estimated position along x axis and y axis.</p> Full article ">Figure 7
<p>Comparison of estimated velocity RMSE between SMM algorithm and adaptive Kalman filter versus different detection error (Scenario 1).</p> Full article ">Figure 8
<p>Comparison of estimated course RMSE between SMM algorithm and adaptive Kalman filter versus different detection error (Scenario 1).</p> Full article ">Figure 9
<p>Comparison of estimated velocity RMSE between SMM algorithm and adaptive Kalman filter versus different detection error (Scenario 7).</p> Full article ">Figure 10
<p>Comparison of estimated course RMSE between SMM algorithm and adaptive Kalman filter versus different detection error (Scenario 7).</p> Full article ">Figure 11
<p>Underwater target tracking and its AOU.</p> Full article ">Figure 12
<p>AOU coverage with 20,000 simulated random points.</p> Full article ">Figure 13
<p>Marginal distribution of AOU.</p> Full article ">Figure 14
<p>RMSE of predicted position versus different prediction time (Scenario 1, <math display="inline"><semantics> <mrow> <msub> <mi>P</mi> <mi>e</mi> </msub> </mrow> </semantics></math> = <math display="inline"><semantics> <mrow> <msub> <mi>P</mi> <mrow> <mi>e</mi> <mn>1</mn> </mrow> </msub> </mrow> </semantics></math>).</p> Full article ">Figure 15
<p>RMSE comparison of predicted position versus different detection error. (<b>a</b>) RMSE of predicted position with seven points. (<b>b</b>) RMSE of predicted position with nine points. (<b>c</b>) RMSE of predicted position with eleven points. (<b>d</b>) RMSE of predicted position with thirteen points.</p> Full article ">Figure 16
<p>AOU semi-major comparison between the SMM algorithm and the adaptive Kalman filter (<math display="inline"><semantics> <mrow> <msub> <mi>P</mi> <mi>e</mi> </msub> </mrow> </semantics></math> = <math display="inline"><semantics> <mrow> <msub> <mi>P</mi> <mrow> <mi>e</mi> <mn>4</mn> </mrow> </msub> </mrow> </semantics></math>).</p> Full article ">Figure 17
<p>AOU semi-minor comparison between the SMM algorithm and the adaptive Kalman filter (<math display="inline"><semantics> <mrow> <msub> <mi>P</mi> <mi>e</mi> </msub> </mrow> </semantics></math> = <math display="inline"><semantics> <mrow> <msub> <mi>P</mi> <mrow> <mi>e</mi> <mn>4</mn> </mrow> </msub> </mrow> </semantics></math>).</p> Full article ">Figure 18
<p>Comparison of estimated AOU size with different measurement errors.</p> Full article ">

Figure 1
<p>The relative motion of the missile and target.</p> Full article ">Figure 2
<p>Dynamics of the missile in the pitch plane.</p> Full article ">Figure 3
<p>The maneuver characteristics of the target.</p> Full article ">Figure 4
<p>Missile–Target pursuit trajectory by using EDO and SMCBS.</p> Full article ">Figure 5
<p>(<b>a</b>) The curve of LOS angular rate; (<b>b</b>) The curve of rudder deflection; (<b>c</b>) The curve of the sliding mode surface proposed in this paper; (<b>d</b>) The curve of the pitch angular rate; (<b>e</b>) The curve of the pitch angle; (<b>f</b>) The curve of the attack angle.</p> Full article ">Figure 5 Cont.
<p>(<b>a</b>) The curve of LOS angular rate; (<b>b</b>) The curve of rudder deflection; (<b>c</b>) The curve of the sliding mode surface proposed in this paper; (<b>d</b>) The curve of the pitch angular rate; (<b>e</b>) The curve of the pitch angle; (<b>f</b>) The curve of the attack angle.</p> Full article ">Figure 6
<p>The maneuver characteristics of the target.</p> Full article ">Figure 7
<p>(<b>a</b>) Missile–Target pursuit trajectory; (<b>b</b>) The curve of the LOS angular rate; (<b>c</b>) The curve of the rudder deflection; (<b>d</b>) The curve of the sliding mode surface; (<b>e</b>) The curve of the pitch angular rate; (<b>f</b>) The curve of the attack angle.</p> Full article ">Figure 7 Cont.
<p>(<b>a</b>) Missile–Target pursuit trajectory; (<b>b</b>) The curve of the LOS angular rate; (<b>c</b>) The curve of the rudder deflection; (<b>d</b>) The curve of the sliding mode surface; (<b>e</b>) The curve of the pitch angular rate; (<b>f</b>) The curve of the attack angle.</p> Full article ">Figure 8
<p>(<b>a</b>) The miss distance in x and y; (<b>b</b>) The characteristic of the relative distance of M—T.</p> Full article ">

Figure 1
<p>The summary framework of the proposed methodology. Source: Own research started from data presented in [<a href="#B2-sensors-22-06670" class="html-bibr">2</a>].</p> Full article ">Figure 2
<p>The baseline model for 100% HDVs and the CAV-adjusted roundabout capacity curves under different penetration rates of CAVs for: (<b>a</b>) the one-lane entry conflicted by one circulating lane (where dashed regression is extrapolated beyond the data); (<b>b</b>) the left entry lane of a two-lane entry conflicted by two circulating lanes (where the operational range depends on the geometry under study). Source: Own research based on data presented in [<a href="#B2-sensors-22-06670" class="html-bibr">2</a>].</p> Full article ">Figure 3
<p>The single-lane roundabout (roundabout 1): (<b>a</b>) the sketch of the roundabout; (<b>b</b>) the network model built in Aimsun Next with labels for the centroids (N: North; S: South; E: East; W: West).</p> Full article ">Figure 4
<p>The two-lane roundabout (roundabout 2): (<b>a</b>) the sketch of the roundabout; (<b>b</b>) the network model built in Aimsun Next with labels for the centroids (N: North; S: South; E: East; W: West).</p> Full article ">Figure 5
<p>Comparisons between the CAV-adjusted and simulated entry capacity data for different market penetration rates of CAVs and HDVs for the entry mechanism 1 (i.e., one-lane entry conflicted by one circulating lane) listed as: (<b>a</b>) baseline scenario: 0% CAVs and 100% HDVs; (<b>b</b>) scenario 1: 20% CAVs and 80% HDVs; (<b>c</b>) scenario 2: 40% CAVs and 60% HVs; (<b>d</b>) scenario 3: 60% CAVs and 40% HDVs; (<b>e</b>) scenario 4: 80% CAVs and 20% HDVs; (<b>f</b>) scenario 5: 100% CAVs and 0% HDVs. Notes: to explain the entry mechanisms 1 reference has been made to the eastbound entry in <a href="#sensors-22-06670-f003" class="html-fig">Figure 3</a>a; the black line and dotted blue line regressions were extrapolated by the data.</p> Full article ">Figure 5 Cont.
<p>Comparisons between the CAV-adjusted and simulated entry capacity data for different market penetration rates of CAVs and HDVs for the entry mechanism 1 (i.e., one-lane entry conflicted by one circulating lane) listed as: (<b>a</b>) baseline scenario: 0% CAVs and 100% HDVs; (<b>b</b>) scenario 1: 20% CAVs and 80% HDVs; (<b>c</b>) scenario 2: 40% CAVs and 60% HVs; (<b>d</b>) scenario 3: 60% CAVs and 40% HDVs; (<b>e</b>) scenario 4: 80% CAVs and 20% HDVs; (<b>f</b>) scenario 5: 100% CAVs and 0% HDVs. Notes: to explain the entry mechanisms 1 reference has been made to the eastbound entry in <a href="#sensors-22-06670-f003" class="html-fig">Figure 3</a>a; the black line and dotted blue line regressions were extrapolated by the data.</p> Full article ">Figure 6
<p>Scattergram analysis to compare the CAV-adjusted and simulated entry capacity data for different proportion of CAVs and HDVs for the entry mechanism 1 (i.e., one-lane entry conflicted by one circulating lane) listed as: (<b>a</b>) baseline scenario: 0% CAVs and 100% HDVs; (<b>b</b>) scenario 1: 20% CAVs and 80% HDVs; (<b>c</b>) scenario 2: 40% CAVs and 60% HVs; (<b>d</b>) scenario 3: 60% CAVs and 40% HDVs; (<b>e</b>) scenario 4: 80% CAVs and 20% HDVs; (<b>f</b>) scenario 5: 100% CAVs and 0% HDVs. Note: to explain the entry mechanisms 1 reference has been made to the eastbound entry in <a href="#sensors-22-06670-f003" class="html-fig">Figure 3</a>a.</p> Full article ">Figure 6 Cont.
<p>Scattergram analysis to compare the CAV-adjusted and simulated entry capacity data for different proportion of CAVs and HDVs for the entry mechanism 1 (i.e., one-lane entry conflicted by one circulating lane) listed as: (<b>a</b>) baseline scenario: 0% CAVs and 100% HDVs; (<b>b</b>) scenario 1: 20% CAVs and 80% HDVs; (<b>c</b>) scenario 2: 40% CAVs and 60% HVs; (<b>d</b>) scenario 3: 60% CAVs and 40% HDVs; (<b>e</b>) scenario 4: 80% CAVs and 20% HDVs; (<b>f</b>) scenario 5: 100% CAVs and 0% HDVs. Note: to explain the entry mechanisms 1 reference has been made to the eastbound entry in <a href="#sensors-22-06670-f003" class="html-fig">Figure 3</a>a.</p> Full article ">Figure 7
<p>Comparisons between the CAV-adjusted and simulated entry capacity data for different market penetration rates of CAVs and HDVs for the entry mechanism 2 (i.e., the left entry lane of two-lane entry approach conflicted by two circulating lanes) listed as: (<b>a</b>) baseline scenario: 0% CAVs and 100% HDVs; (<b>b</b>) scenario 1: 20% CAVs and 80% HDVs; (<b>c</b>) scenario 2: 40% CAVs and 60% HVs; (<b>d</b>) scenario 3: 60% CAVs and 40% HDVs; (<b>e</b>) scenario 4: 80% CAVs and 20% HDVs; (<b>f</b>) scenario 5: 100% CAVs and 0% HDVs. Notes: to explain the entry mechanisms 2 reference has been made to the northbound left entry lane entry in <a href="#sensors-22-06670-f004" class="html-fig">Figure 4</a>a; the black line and dotted red line regressions were extrapolated by the data.</p> Full article ">Figure 7 Cont.
<p>Comparisons between the CAV-adjusted and simulated entry capacity data for different market penetration rates of CAVs and HDVs for the entry mechanism 2 (i.e., the left entry lane of two-lane entry approach conflicted by two circulating lanes) listed as: (<b>a</b>) baseline scenario: 0% CAVs and 100% HDVs; (<b>b</b>) scenario 1: 20% CAVs and 80% HDVs; (<b>c</b>) scenario 2: 40% CAVs and 60% HVs; (<b>d</b>) scenario 3: 60% CAVs and 40% HDVs; (<b>e</b>) scenario 4: 80% CAVs and 20% HDVs; (<b>f</b>) scenario 5: 100% CAVs and 0% HDVs. Notes: to explain the entry mechanisms 2 reference has been made to the northbound left entry lane entry in <a href="#sensors-22-06670-f004" class="html-fig">Figure 4</a>a; the black line and dotted red line regressions were extrapolated by the data.</p> Full article ">Figure 8
<p>Scattergram analysis to compare the CAV-adjusted versus simulated entry capacity data for different proportion of CAVs and HDVs for the entry mechanism 2 (i.e., the left entry lane of two-lane entry approach conflicted by two circulating lanes) listed as: (<b>a</b>) baseline scenario: 0% CAVs and 100% HDVs; (<b>b</b>) scenario 1: 20% CAVs and 80% HDVs; (<b>c</b>) scenario 2: 40% CAVs and 60% HDVs; (<b>d</b>) scenario 3: 60% CAVs and 40% HDVs; (<b>e</b>) scenario 4: 80% CAVs and 20% HDVs; (<b>f</b>) scenario 5: 100% CAVs and 0% HDVs. Note: to explain the entry mechanisms 2 reference has been made to the northbound left entry lane entry in <a href="#sensors-22-06670-f004" class="html-fig">Figure 4</a>a.</p> Full article ">Figure 8 Cont.
<p>Scattergram analysis to compare the CAV-adjusted versus simulated entry capacity data for different proportion of CAVs and HDVs for the entry mechanism 2 (i.e., the left entry lane of two-lane entry approach conflicted by two circulating lanes) listed as: (<b>a</b>) baseline scenario: 0% CAVs and 100% HDVs; (<b>b</b>) scenario 1: 20% CAVs and 80% HDVs; (<b>c</b>) scenario 2: 40% CAVs and 60% HDVs; (<b>d</b>) scenario 3: 60% CAVs and 40% HDVs; (<b>e</b>) scenario 4: 80% CAVs and 20% HDVs; (<b>f</b>) scenario 5: 100% CAVs and 0% HDVs. Note: to explain the entry mechanisms 2 reference has been made to the northbound left entry lane entry in <a href="#sensors-22-06670-f004" class="html-fig">Figure 4</a>a.</p> Full article ">Figure 9
<p>Percentage differences in the “what-if” scenarios compared to the baseline scenario with HDVs only: (<b>a</b>) entry capacity, (<b>b</b>) delay and (<b>c</b>) travel time. Note: (1) 20% CAVs vs. baseline scenario; (2) 40% CAVs vs. baseline scenario; (3) 60% CAVs vs. baseline scenario; (4) 80% CAVs vs. baseline scenario; (5) 100% CAVs vs. baseline scenario.</p> Full article ">

Graphical abstract
Full article ">Figure 1
<p>Flow diagram of the proposed deep-learning-based multiple model tracking method.</p> Full article ">Figure 2
<p>Structure of the adaptive turn rate estimation network.</p> Full article ">Figure 3
<p>Bi-LSTM structure diagram.</p> Full article ">Figure 4
<p>Structure of the filter state modification network.</p> Full article ">Figure 5
<p>Flowchart of dataset construction.</p> Full article ">Figure 6
<p>Single simulation results for different maneuvering modes: (<b>a</b>) tracking results, (<b>b</b>) turn rate estimates, and (<b>c</b>) tracking errors.</p> Full article ">Figure 6 Cont.
<p>Single simulation results for different maneuvering modes: (<b>a</b>) tracking results, (<b>b</b>) turn rate estimates, and (<b>c</b>) tracking errors.</p> Full article ">Figure 7
<p>Tracking error comparisons of test datasets: (<b>a</b>) group 1, (<b>b</b>) group 2, (<b>c</b>) group 3, (<b>d</b>) group 4, and (<b>e</b>) group 5.</p> Full article ">Figure 7 Cont.
<p>Tracking error comparisons of test datasets: (<b>a</b>) group 1, (<b>b</b>) group 2, (<b>c</b>) group 3, (<b>d</b>) group 4, and (<b>e</b>) group 5.</p> Full article ">Figure 8
<p>Simulation scenario. (△—start position, ◯—stop position).</p> Full article ">Figure 9
<p>OSPA distance comparison.</p> Full article ">Figure 10
<p>RMSE of turn rate estimates.</p> Full article ">Figure 11
<p>Tracking results of a single simulation.</p> Full article ">Figure 12
<p>Result comparisons of turn rate estimate: (<b>a</b>) target 1, (<b>b</b>) target 2, (<b>c</b>) target 3, (<b>d</b>) target 4, and (<b>e</b>) target 5.</p> Full article ">Figure 12 Cont.
<p>Result comparisons of turn rate estimate: (<b>a</b>) target 1, (<b>b</b>) target 2, (<b>c</b>) target 3, (<b>d</b>) target 4, and (<b>e</b>) target 5.</p> Full article ">Figure 13
<p>Experimental scenario.</p> Full article ">Figure 14
<p>Comparisons of real UAV tracking results: (<b>a</b>) GMPHD, (<b>b</b>) GMPHD-ATN, and (<b>c</b>) GMPHD-ATN-FMN.</p> Full article ">

Figure 1
<p>High-squint SAR geometry with a ground maneuvering moving target.</p> Full article ">Figure 2
<p>Range: Please add bold for abc in figure, same as others. cell migration and azimuth phase comparison. (<b>a</b>) Linear range walk difference. (<b>b</b>) Quadratic phase difference. (<b>c</b>) Third-order phase difference.</p> Full article ">Figure 3
<p>Flowchart of the proposed HPC-GMTIm algorithm.</p> Full article ">Figure 4
<p>Subaperture processing results. The whole aperture is divided into 8 subapertures, and all 8 subaperture images are well focused by the proposed HPC-GMTIm algorithm.</p> Full article ">Figure 5
<p>Whole aperture image. (<b>a</b>) HPC-GMTIm algorithm. (<b>b</b>) GHHAF method. (<b>c</b>) Cross-range profile comparison.</p> Full article ">Figure 6
<p>Moving target detection for T1. (<b>a</b>) Range–Doppler domain. (<b>b</b>) Moving target detection result.</p> Full article ">Figure 7
<p>The stationary scene image after eliminating the energy of T1.</p> Full article ">Figure 8
<p>The envelope of T1. (<b>a</b>) Before correction. (<b>b</b>) After correction.</p> Full article ">Figure 9
<p>The imaging result of T1. (<b>a</b>) HPC-GMTIm (entropy = 2.97). (<b>b</b>) GHHAF method (entropy = 3.36).</p> Full article ">Figure 10
<p>The stationary scene image containing moving target T2.</p> Full article ">Figure 11
<p>Moving target detection for T2. (<b>a</b>) Subaperture images before detection. (<b>b</b>) Subaperture images after detection.</p> Full article ">Figure 12
<p>The envelope of T2. (<b>a</b>) Before correction. (<b>b</b>) After correction.</p> Full article ">Figure 13
<p>The imaging result of T2. (<b>a</b>) HPC-GMTIm (entropy = 3.62). (<b>b</b>) GHHAF method (entropy = 4.17).</p> Full article ">Figure 14
<p>Moving target detection for T3. (<b>a</b>) Range–Doppler domain. (<b>b</b>) Moving target detection result.</p> Full article ">Figure 15
<p>The stationary scene image corresponding to T3.</p> Full article ">Figure 16
<p>The envelope of T3. (<b>a</b>) Before correction. (<b>b</b>) After correction.</p> Full article ">Figure 17
<p>The imaging result of T3. (<b>a</b>) HPC-GMTIm (entropy = 2.48). (<b>b</b>) GHHAF method (entropy = 2.81).</p> Full article ">Figure 18
<p>The stationary scene containing moving target T4.</p> Full article ">Figure 19
<p>Moving target detection for T4. (<b>a</b>) Subaperture images before detection. (<b>b</b>) Subaperture images after detection.</p> Full article ">Figure 20
<p>The envelope of T4. (<b>a</b>) Before RCMC. (<b>b</b>) After RCMC.</p> Full article ">Figure 21
<p>The imaging result of T4. (<b>a</b>) HPC-GMTIm (entropy = 3.46). (<b>b</b>) GHHAF method (entropy = 4.03).</p> Full article ">

Figure 1
<p>Obesity and oxidative stress. Cell intrinsic and cell extrinsic factors related to obesity, including inflammation, hyperglycemia, hyperleptinemia, hyperlipidemia, reduced adiponectin levels, obstructive sleep apnea (OSA), or mitochondria-associated Endoplasmic Reticulum (ER) membranes (MAMs) defects, can exacerbate the production of reactive oxygen species (ROS). ROS account for the induction of a detrimental lipogenic response (dependent on SREBP1 and fatty acid synthase (FAS)), which contributes to, or worsens, the obese phenotype.</p> Full article ">Figure 2
<p>Molecular mechanisms of autophagy. In response to diverse stimuli, the activation of AMPK and/or the inhibition of mTORC1 stimulate the function of the ULK1 complex, which in turn activates the BECN1/VPS34 complex, starting the formation of the phagophore. Several ATG proteins catalyze the conjugation of the cargo adaptor LC3 to PE residues on the expanding phagophore membrane. Mature double membrane autophagosome loaded with its cargo eventually undergoes fusion with the lysosome, whereby its content is degraded and recycled. AMPK, 5′-adenosine monophosphate (AMP)-activated protein kinase; ATG, autophagy-related proteins; BECN1, beclin-1; CAV1, Caveolin 1, ER, endoplasmic reticulum; LC3, light chain of protein 1 associated to microtubules 3; MTOR, mechanistic target of rapamycin; PE, phosphatidylethanolamine; SQSTM1, sequestosome 1 protein (p62); ULK, UNC-51–like kinase.</p> Full article ">Figure 3
<p>Reactive oxygen species–mediated autophagy induction. Autophagy is induced by ROS at different levels across the machinery, including initiation (through direct or indirect ROS-mediated modulation of AMPK and mTORC1 activity); nucleation, via Caveolin-1 or PKD-dependent activation of the VPS34/BECN1 complex and elongation, through ROS-dependent activation of Atg4. KEAP1. AMPK, 5′-adenosine monophosphate (AMP)-activated protein kinase; ATG, autophagy related proteins; ATM, ataxia-telangiectasia mutated; BECN1, beclin-1; ER, endoplasmic reticulum; KEAP1, Kelch ECH associating protein 1; LC3, light chain of protein 1 associated to microtubules 3; MTOR, mechanistic target of rapamycin; NRF2, nuclear factor erythroid 2-related factor 2; PE, phosphatidylethanolamine; PKD, protein kinase D; PTEN, phosphatase and tensin homolog; SQSTM1, sequestosome 1 protein (p62); ULK, UNC-51–like kinase.</p> Full article ">

Figure 1
<p>The processing result of a typical piece of data. (<b>a</b>) is the time-range image of the typical data; (<b>b</b>) is the time-velocity image of the target after short-time Fourier transform</p> Full article ">Figure 2
<p>The histogram of acceleration of 4600 insects.</p> Full article ">Figure 3
<p>The flowchart of the integration algorithm based on dynamic programming and fractional Fourier transforming (DP-FrFT).</p> Full article ">Figure 4
<p>The integration gain in different methods. (<b>a</b>) is the integration gain with the number of integrated pulse in moving target detector (MTD); (<b>b</b>) is the result in Keystone-MTD; (<b>c</b>) is the result in fractional Fourier transforming (FrFT); (<b>d</b>) is the result in DP-FrFT (our method).</p> Full article ">Figure 5
<p>Time-consuming of different methods.</p> Full article ">Figure 6
<p>Detection probability of four methods under different signal-to-noise ratios (SNRs).</p> Full article ">Figure 7
<p>The motion parameters estimation errors of two methods. (<b>a</b>) shows the different speed error with SNR in our method and Drake’s; (<b>b</b>) shows the different acceleration error with SNR in our method and Drake’s.</p> Full article ">Figure 8
<p>The radar system and its working environment. (<b>a</b>) shows the Ku-band vertical-beam entomological radar; (<b>b</b>) shows the working environment of the radar.</p> Full article ">Figure 9
<p>The typical detection result by MTD and DP-FrFT. (<b>a</b>) is the time-range image corresponding to the detection result; (<b>b</b>) is the detection results by MTD; (<b>c</b>) is the detection results by DP-FrFT.</p> Full article ">

Figure 1
<p>The architecture of our proposed M³C tracking by detection pipeline. It comprises eight modules, among which the video stabilization unit uses the algorithm introduced by [<a href="#B33-electronics-08-00723" class="html-bibr">33</a>]. We proposed the four blue units, and their interaction with the other three green units is discussed in detail later.</p> Full article ">Figure 2
<p>Framework of the detector. Among them, Darknetconv2d_BN_Leaky (DBL) is the basic unit, which consists of a convolutional layer (conv), batch normalization, and a leaky rectified linear unit (ReLU).</p> Full article ">Figure 3
<p>Typical structure of a gated recurrent unit (GRU) neural network. <math display="inline"> <semantics> <mrow> <msub> <mi>O</mi> <mi>t</mi> </msub> <mo stretchy="false">(</mo> <msub> <mi>h</mi> <mi>t</mi> </msub> <mo stretchy="false">)</mo> </mrow> </semantics> </math> is the output at time <math display="inline"> <semantics> <mi>t</mi> </semantics> </math>; <math display="inline"> <semantics> <mrow> <msub> <mi>r</mi> <mi>t</mi> </msub> </mrow> </semantics> </math> denotes the reset gate, which determines the combination of current input and historical memory information; and, <math display="inline"> <semantics> <mrow> <msub> <mi>z</mi> <mi>t</mi> </msub> </mrow> </semantics> </math> is the update gate, which determines the proportion of memory left behind.</p> Full article ">Figure 4
<p>The improved framework of the GRU encoder–decoder with attention. The attention model was embedded between encoders and decoders and learned the attention weights <math display="inline"> <semantics> <mrow> <msub> <mi>c</mi> <mrow> <mi>t</mi> <mo>−</mo> <mi>N</mi> <mi>τ</mi> <mo> </mo> </mrow> </msub> <mo>,</mo> <mo>…</mo> <mo>,</mo> <msub> <mi>c</mi> <mrow> <mi>t</mi> <mo>−</mo> <mi>τ</mi> </mrow> </msub> <mo>,</mo> <msub> <mi>c</mi> <mi>t</mi> </msub> <mo>,</mo> <msub> <mi>c</mi> <mrow> <mi>t</mi> <mo>+</mo> <mi>τ</mi> </mrow> </msub> </mrow> </semantics> </math> via the fully-connected network (FC) and softmax with loss function, output vector of encoder <math display="inline"> <semantics> <mrow> <msub> <mi>h</mi> <mrow> <mi>t</mi> <mo>−</mo> <mi>N</mi> <mi>τ</mi> <mo> </mo> </mrow> </msub> <mo>,</mo> <mo>…</mo> <mo>,</mo> <msub> <mi>h</mi> <mrow> <mi>t</mi> <mo>−</mo> <mi>τ</mi> </mrow> </msub> <mo>,</mo> <msub> <mi>h</mi> <mi>t</mi> </msub> <mo>,</mo> <msub> <mi>h</mi> <mrow> <mi>t</mi> <mo>+</mo> <mi>τ</mi> </mrow> </msub> </mrow> </semantics> </math>, and state vector of decoder <math display="inline"> <semantics> <mrow> <msub> <mi>s</mi> <mrow> <mi>t</mi> <mo>−</mo> <mo stretchy="false">(</mo> <mi>N</mi> <mo>+</mo> <mn>1</mn> <mo stretchy="false">)</mo> <mi>τ</mi> </mrow> </msub> <mo>,</mo> <msub> <mi>s</mi> <mrow> <mi>t</mi> <mo>−</mo> <mi>N</mi> <mi>τ</mi> </mrow> </msub> <mo>,</mo> <mo>…</mo> <mo>,</mo> <msub> <mi>s</mi> <mrow> <mi>t</mi> <mo>−</mo> <mi>τ</mi> </mrow> </msub> <mo>,</mo> <msub> <mi>s</mi> <mi>t</mi> </msub> </mrow> </semantics> </math> as the attention model’s input sequence.</p> Full article ">Figure 5
<p>Sketch map of adaptive association gate of appearance.</p> Full article ">Figure 6
<p>Visual tracking results of video sequence (MVI_1448_VIS_Haze) with occlusion. Six vessels have been stably tracked in frame #0126 (<b>a</b>). Starting in frame #0264 (<b>b</b>), Sb001 was occluded by Tug001 until being completely occluded in frame #0355 (<b>c</b>) and then reappeared in frame #0370 until it was completely visible in frame #0438 as shown in (<b>d</b>).</p> Full article ">Figure 6 Cont.
<p>Visual tracking results of video sequence (MVI_1448_VIS_Haze) with occlusion. Six vessels have been stably tracked in frame #0126 (<b>a</b>). Starting in frame #0264 (<b>b</b>), Sb001 was occluded by Tug001 until being completely occluded in frame #0355 (<b>c</b>) and then reappeared in frame #0370 until it was completely visible in frame #0438 as shown in (<b>d</b>).</p> Full article ">Figure 7
<p>Visual tracking results of video sequence (MVI_0799_VIS_OB) with reappear after lost. In frame #0075 (<b>a</b>), a vessel has been stably tracked. Due to the rolling of the ego vessel, Passenger001 gradually disappeared from the camera scene in frame #0076 until frame #0078 (<b>b</b>), when it completely disappeared. Subsequently, it began to reappear in frame #0091 until frame #0094 (<b>c</b>), when it was completely visible. As shown in frame #0095 (<b>d</b>), the vessel was re-tracked steadily.</p> Full article ">

Figure 1
<p>Interacting multiple model (IMM) algorithm principle frame.</p> Full article ">Figure 2
<p>Membership functions of (<b>a</b>) <b>EoR</b> and (<b>b</b>) Δ<b>R</b>.</p> Full article ">Figure 3
<p>Structure of the single-node fuzzy neural network (FNN) inference machine.</p> Full article ">Figure 4
<p>Principle frame graph of FNN fusion system (FNNFS).</p> Full article ">Figure 5
<p>Decision functions of (<b>a</b>) <b>EoR</b> and (<b>b</b>) <b>R</b>.</p> Full article ">Figure 6
<p>Structure of the multi-node FNN.</p> Full article ">Figure 7
<p>Detected target trajectory of the net nodes. (<b>a</b>) Detected trajectory of node 1; (<b>b</b>) Detected trajectory of node 2; (<b>c</b>) Detected trajectory of node 3; (<b>d</b>) Detected trajectory of node 4.</p> Full article ">Figure 8
<p>Computational results of the FNN–IMM. (<b>a</b>) Position error of each node. (<b>b</b>) Estimated trajectory of Nodes 1, 2, and 3 and the network output.</p> Full article ">Figure 9
<p>Computational results of FNN–IMM, IMM-UKF, IMM-EKF, and VB-IMM. (<b>a</b>) Mean trajectory after 100 Monte Carlo trials. (<b>b</b>) Estimation errors of target position by FNN-IMM, IMM-UKF, IMM-EKF, and VB-IMM.</p> Full article ">Figure 10
<p>Target positional errors of the network outputs and Node 3.</p> Full article ">

<p>The shape of function <span class="html-italic">φ</span>(<span class="html-italic">z</span>) with <span class="html-italic">σ</span> = 1.</p> Full article ">
<p>Plot of estimated results generated from a single run of the different filters with non-stationary observation model.</p> Full article ">
<p>Errors of different filters versus the process noises in a single run.</p> Full article ">
<p>The average RMSEs of three algorithms calculated over 100 independent runs with the linear and non-linear observation model.</p> Full article ">
<p>The average execution time of three particle filters for 60 steps tracking over 100 independent runs.</p> Full article ">