Method, device, electronic device, storage medium and computer program product for evaluating adsorbability

Abstract

The present disclosure relates to an evaluation method and device for adsorbability, electronic equipment, storage medium and computer program product, comprising: emitting a pulse signal to a target to be adsorbed through a multi-zone ToF sensor; receiving a reflected pulse signal through the multi-zone ToF sensor; inputting the reflected pulse signal into a time domain branch to obtain a time domain feature; inputting the reflected pulse signal into a time-frequency domain branch to obtain a time-frequency domain feature; fusing the time domain feature and the time-frequency domain feature to obtain a fused feature; and evaluating the adsorbability of the target to be adsorbed based on the fused feature. In this way, by setting a multi-zone ToF sensor as a core perception unit on a bionic underwater robot fish, a multi-dimensional input with spatial geometric information and time dynamic characteristics can be constructed. In addition, by co-modeling the multi-dimensional input through the parallel time domain branch and time-frequency domain branch included in the space-time perception model, the evaluation accuracy of the host adsorbability can be improved.

Description

Technical Field

[0001] This disclosure relates to the field of robotics, and more specifically, to methods, apparatus, electronic devices, storage media, and computer program products for evaluating adsorption properties. Background Technology

[0002] Biomimetic underwater robotic fish exhibit significant advantages in hydrodynamic efficiency, maneuverability, and ecological compatibility. However, the endurance of small biomimetic underwater robotic fish remains limited, due to their constrained size and payload capacity. In nature, remoras have evolved unique sucker structures on their backs, allowing them to attach to the surfaces of larger hosts and effectively expand their range of motion. Inspired by this, the host adaptability of biomimetic underwater robotic fish can be assessed through their proprioceptive capabilities, thereby enabling them to expand their range of motion by attaching to hosts.

[0003] However, in near-field scenarios, the reliability of visual perception is significantly reduced due to limitations in viewing angle, severe occlusion, and decreased depth estimation accuracy. These constraints hinder the accurate acquisition of target surface curvature and local motion cues. Therefore, an effective adsorption system must integrate a suitable sensing mechanism to guide the selection of adsorption location and timing. While technologies have introduced single-point Time-of-Flight (ToF) sensors into adsorption operations, these sensors struggle to capture local geometric changes on the target surface, resulting in poor accuracy in assessing the host's adsorption properties. Summary of the Invention

[0004] This disclosure provides methods, apparatus, electronic devices, storage media, and computer program products for evaluating adsorption properties, in order to at least solve the problem in the aforementioned related technologies that single-point ToF sensors have difficulty capturing local geometric feature changes on the target surface, resulting in poor accuracy in evaluating the host's adsorption properties.

[0005] According to a first aspect of the present disclosure, a method for evaluating adsorption properties is provided, applied to a biomimetic underwater robotic fish. The biomimetic underwater robotic fish is equipped with a multi-zone Time-of-Flight (ToF) sensor and a spatiotemporal sensing model. The spatiotemporal sensing model includes a time-domain branch and a time-frequency domain branch. The evaluation method includes: transmitting a pulse signal to a target to be adsorbed via the multi-zone ToF sensor; receiving a reflected pulse signal obtained by the multi-zone ToF sensor after the pulse signal is reflected by the target to be adsorbed; inputting the reflected pulse signal into the time-domain branch to obtain time-domain features, wherein the time-domain features characterize the amplitude variation of the reflected pulse signal over time; inputting the reflected pulse signal into the time-frequency domain branch to obtain time-frequency domain features, wherein the time-frequency domain features characterize the frequency variation of the reflected pulse signal over time; fusing the time-domain features and the time-frequency domain features to obtain fused features; and evaluating the adsorption properties of the target to be adsorbed based on the fused features.

[0006] Optionally, the temporal branch includes a two-layer residual convolutional block, a global average pooling module, a fully connected module, and a bidirectional long short-term memory network; the step of inputting the reflected pulse signal into the temporal branch to obtain temporal features includes: inputting the reflected pulse signal into the two-layer residual convolutional block to obtain local spatial features; inputting the local spatial features into the global average pooling module to obtain global average pooling features; inputting the global average pooling features into the fully connected module to obtain fully connected features; and inputting the fully connected features into the bidirectional long short-term memory network to capture long-term temporal dependencies as the temporal features.

[0007] Optionally, the time-frequency domain branch includes a short-time Fourier transform module, a convolutional neural network, an adaptive average pooling module, and a fully connected layer; the step of inputting the reflected pulse signal into the time-frequency domain branch to obtain time-frequency domain features includes: inputting the reflected pulse signal into the short-time Fourier transform module to obtain a time-frequency graph; inputting the time-frequency graph into the convolutional neural network to obtain local time-frequency features; inputting the local time-frequency features into the adaptive average pooling module to obtain adaptive average pooling features; and inputting the adaptive average pooling features into the fully connected layer to obtain the time-frequency domain features.

[0008] Optionally, the adsorption properties include the surface geometry of the target to be adsorbed and / or the motion characteristics of the target to be adsorbed.

[0009] Optionally, the surface geometry properties include the following types: plane, curved surface.

[0010] Optionally, the motion characteristics include vibration type and / or vibration frequency, wherein the vibration type is divided into translational vibration and rotational vibration.

[0011] According to a second aspect of the present disclosure, an evaluation device for adsorption properties is provided, applied to a biomimetic underwater robotic fish. The biomimetic underwater robotic fish is equipped with a multi-zone ToF sensor and a spatiotemporal sensing model. The spatiotemporal sensing model includes a time-domain branch and a time-frequency domain branch. The evaluation device includes: a signal transmitting module configured to transmit a pulse signal to a target to be adsorbed via the multi-zone ToF sensor; a signal receiving module configured to receive a reflected pulse signal obtained by the reflection of the pulse signal by the target to be adsorbed via the multi-zone ToF sensor; and a time-domain feature acquisition module configured to... The reflected pulse signal is input into the time-domain branch to obtain time-domain features, wherein the time-domain features are used to characterize the amplitude variation of the reflected pulse signal over time; the time-frequency domain feature acquisition module is configured to input the reflected pulse signal into the time-frequency domain branch to obtain time-frequency domain features, wherein the time-frequency domain features are used to characterize the frequency variation of the reflected pulse signal over time; the fusion module is configured to fuse the time-domain features and the time-frequency domain features to obtain fused features; the characteristic evaluation module is configured to evaluate the adsorption characteristics of the target to be adsorbed based on the fused features.

[0012] Optionally, the temporal branch includes a two-layer residual convolutional block, a global average pooling module, a fully connected module, and a bidirectional long short-term memory network; the temporal feature acquisition module is configured to: input the reflected pulse signal into the two-layer residual convolutional block to obtain local spatial features; input the local spatial features into the global average pooling module to obtain global average pooling features; input the global average pooling features into the fully connected module to obtain fully connected features; and input the fully connected features into the bidirectional long short-term memory network to capture long-term temporal dependencies as the temporal features.

[0013] Optionally, the time-frequency domain branch includes a short-time Fourier transform module, a convolutional neural network, an adaptive average pooling module, and a fully connected layer; the time-frequency domain feature acquisition module is configured to: input the reflected pulse signal into the time-frequency domain branch to obtain time-frequency domain features, including: inputting the reflected pulse signal into the short-time Fourier transform module to obtain a time-frequency graph; inputting the time-frequency graph into the convolutional neural network to obtain local time-frequency features; inputting the local time-frequency features into the adaptive average pooling module to obtain adaptive average pooling features; and inputting the adaptive average pooling features into the fully connected layer to obtain the time-frequency domain features.

[0014] Optionally, the adsorption properties include the surface geometry of the target to be adsorbed and / or the motion characteristics of the target to be adsorbed.

[0015] Optionally, the surface geometry properties include the following types: plane, curved surface.

[0016] Optionally, the motion characteristics include vibration type and / or vibration frequency, wherein the vibration type is divided into translational vibration and rotational vibration.

[0017] According to a third aspect of the present disclosure, an electronic device is provided, comprising: a processor; a memory for storing processor-executable instructions; wherein the processor is configured to execute the instructions to implement a method for evaluating adsorption properties according to the present disclosure.

[0018] According to a fourth aspect of the present disclosure, a computer-readable storage medium is provided that, when instructions in the computer-readable storage medium are executed by a processor of an electronic device, enables the electronic device to perform an evaluation method for absorbent properties according to the present disclosure.

[0019] According to a fifth aspect of the present disclosure, a computer program product is provided, including a computer program that, when executed by a processor, implements a method for evaluating absorbability properties according to the present disclosure.

[0020] The technical solutions provided by the embodiments of this disclosure have at least the following beneficial effects: In this disclosure, by equipping a biomimetic underwater robotic fish with a multi-zone Time-of-Flight (ToF) sensor as the core sensing unit, the reflected pulse signals from the target to be adsorbed can be continuously acquired, thereby constructing a multi-dimensional input that combines spatial geometric information and temporal dynamic characteristics. Furthermore, by collaboratively modeling this multi-dimensional input through parallel time-domain and time-frequency domain branches included in the spatiotemporal sensing model, complementary characterization of temporal evolution patterns and spectral characteristics can be achieved. This enables joint sensing of the adsorbability characteristics of the target, improving the accuracy of the assessment of the host's adsorbability characteristics.

[0021] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0022] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure, and are not intended to unduly limit this disclosure.

[0023] Figure 1 This is a schematic diagram illustrating near-field adsorption target sensing according to an exemplary embodiment of the present disclosure; Figure 2 This is a schematic diagram illustrating an evaluation scenario of a ToF sensor according to an exemplary embodiment of the present disclosure; Figure 3This is a schematic diagram illustrating the evaluation results of a ToF sensor according to an exemplary embodiment of the present disclosure; Figure 4 This is a flowchart illustrating a method for evaluating adsorption properties according to exemplary embodiments of the present disclosure; Figure 5 This is a schematic diagram illustrating the time-domain branch and time-frequency domain branch included in a spatiotemporal perception model according to an exemplary embodiment of the present disclosure; Figure 6 This is a schematic diagram showing a comparison between the estimated distributions of a single modality in related technologies and the fusion spatiotemporal perception model provided in this disclosure for six typical motion types; Figure 7 This is a graph showing the frequency estimation results of the spatiotemporal perception model according to an exemplary embodiment of the present disclosure in a real-world scenario under different target distances and dynamic frequency variations; Figure 8 This is a block diagram illustrating an evaluation apparatus for adsorption properties according to exemplary embodiments of the present disclosure; Figure 9 This is a block diagram illustrating an electronic device according to exemplary embodiments of the present disclosure. Detailed Implementation

[0024] To enable those skilled in the art to better understand the technical solutions of this disclosure, the technical solutions in the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings.

[0025] It should be noted that the terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following examples do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure as detailed in the appended claims.

[0026] It should be noted that the phrase "at least one of several items" in this disclosure refers to three parallel cases: "any one of the several items", "a combination of any number of the several items", and "all of the several items". For example, "including at least one of A and B" includes the following three parallel cases: (1) including A; (2) including B; (3) including A and B. Another example is "performing at least one of step one and step two", which means the following three parallel cases: (1) performing step one; (2) performing step two; (3) performing both step one and step two.

[0027] Research shows that miniature multi-zone ToF sensors can be effectively applied to dynamic obstacle avoidance and object detection. Therefore, relying on the sensing information of this type of sensor to characterize the geometric structure and motion dynamics of the target surface is expected to further improve the system's decision-making capabilities. Figure 1 This is a schematic diagram illustrating near-field adsorption target sensing according to an exemplary embodiment of the present disclosure.

[0028] Reference Figure 1 This disclosure employs a multi-region Time-of-Flight (ToF) model to construct a spatiotemporal sensing model, thereby estimating the typical surface curvature and motion characteristics of the adsorbed object. Firstly, this disclosure evaluates the measurement characteristics of the ToF sensor in a typical underwater scenario, providing a platform foundation for near-field sensing.

[0029] Given the advantages of the VL53L7CX ToF sensor, including multi-area sensing, wide field of view, high sampling rate, compact size, and low power consumption, and its widespread application in near-field state perception scenarios for small robots, this disclosure selects this sensor model as the core sensing unit. It should be noted that electromagnetic wave attenuation is significant in underwater environments; therefore, this sensor is primarily suited for the short-range spatial measurement scenarios required by this disclosure. To further verify its practical applicability and performance in underwater environments, this disclosure has conducted targeted evaluation experiments to demonstrate its underwater ranging accuracy and robustness.

[0030] Figure 2 This is a schematic diagram illustrating an evaluation scenario of a ToF sensor according to an exemplary embodiment of the present disclosure. (Refer to...) Figure 2 The evaluation scenario for the VL53L7CX sensor constructed in this disclosure can be as follows: The sensor can be fixedly mounted on a linear guide rail, and then the entire sensor can be immersed in an underwater environment at a depth of 50 mm. The sensor's detection direction needs to be perpendicular to a white acrylic plate (as the standard detection target). Furthermore, to ensure the consistency of measurement data, the target surface within the sensor's field of view needs to remain flat throughout the experiment. During the experiment, the target object can be controlled to move along the linear guide rail in 20 mm increments, with the movement range set from 20 mm to 120 mm. At each stationary position, 1000 frames of 8×8 multi-region distance measurement data can be continuously recorded. The above data is mainly used for subsequent statistical analysis of the sensor's spatial measurement accuracy, noise characteristics, and measurement stability.

[0031] Figure 3 This is a schematic diagram illustrating the evaluation results of a ToF sensor according to an exemplary embodiment of the present disclosure.

[0032] Reference Figure 3The figures present the mean and standard deviation of distance measurement data from a C8×R8 (8×8) multi-region sensor at a distance of 40 mm from the target. The evaluation results show that the sensor exhibits high temporal stability, with the standard deviation of its sensing matrix uniformly distributed within approximately 0.39 mm. However, the sensor's average ranging error shows a non-uniform spatial distribution: the central region (corresponding to the near-normal incident angle and with the least optical distortion) has the highest measurement accuracy, with an average error of approximately 8 mm; in contrast, the corner regions show significantly larger measurement deviations, with an average error approximately 8-10 mm higher than the central region.

[0033] To further comprehensively evaluate the sensor's performance at different distances, this disclosure allows for the sensor to be controlled to move gradually along a linear guide rail, and allows for the selection of measurement data from the four central regions of the sensor for focused analysis. In addition to the original distance matrix, this disclosure also simultaneously records the sensor's built-in validity matrix (which identifies the reliability of each distance data entry). Experimental results show that the sensor's measurement error increases moderately with distance, but remains within an acceptable range overall. The above series of evaluation results fully demonstrate that the VL53L7CX sensor exhibits good robustness and reliability for near-field sensing in underwater environments, and can serve as the core sensing unit of the spatiotemporal sensing framework of this disclosure.

[0034] Figure 4 This is a flowchart illustrating an evaluation method for adsorption properties according to exemplary embodiments of the present disclosure, applied to a biomimetic underwater robotic fish. The biomimetic underwater robotic fish may be equipped with a multi-zone ToF sensor and a spatiotemporal sensing model, which may include time-domain and time-frequency domain branches.

[0035] Reference Figure 4 In step 401, a pulse signal can be emitted to the target to be adsorbed via a multi-zone ToF sensor. In an underwater environment, the biomimetic underwater robotic fish is usually stationary, while the target to be adsorbed can be stationary or moving. For example, the biomimetic underwater robotic fish can be, but is not limited to, a small robotic fish with a length of approximately 60 cm; the target to be adsorbed can be, but is not limited to, a dolphin model with a length of approximately 1.8 meters.

[0036] In step 402, the reflected pulse signal obtained by the pulse signal being reflected by the target to be adsorbed can be received by a multi-zone ToF sensor.

[0037] Figure 5 This is a schematic diagram illustrating the time-domain branch and time-frequency domain branch included in a spatiotemporal perception model according to an exemplary embodiment of the present disclosure. (Refer to...) Figure 5The time-domain branch can include components such as Residual CNNs and Bidirectional Long Short-Term Memory (BiLSTM) networks. A Residual CNN can consist of a two-layer Residual Convolution Block (RCB), a Global Average Pooling (GAP) module, and a Fully Connected (FC) module. The time-frequency domain branch can include components such as Short-Time Fourier Transform (STFT), Convolutional Neural Networks (CNNs), Adaptive Average Pooling (AAP) modules, and fully connected layers.

[0038] In step 403, the reflected pulse signal can be input into the time domain branch to obtain time domain features, which can be used to characterize the variation of the amplitude of the reflected pulse signal over time.

[0039] According to exemplary embodiments of this disclosure, as previously described, the temporal branch may include a two-layer residual convolutional block, a global average pooling module, a fully connected module, and a bidirectional long short-term memory network.

[0040] First, the reflected pulse signal can be input into a two-layer residual convolutional block to obtain local spatial features. Then, the local spatial features can be input into a global average pooling module to obtain global average pooling features. Next, the global average pooling features can be input into a fully connected module to obtain fully connected features. Finally, the fully connected features can be input into a bidirectional long short-term memory network to capture long-term temporal dependencies as temporal features.

[0041] It should be noted that the temporal branch can capture temporal dynamics and local spatial correlations from the original ToF depth frame. Each frame input... ( First, the input is processed through two layers of residual convolutional blocks to extract local spatial features. This input for each frame... The VL53L7CX sensor can obtain the photon arrival time distribution based on the reflected pulse signals received in each grid region (H×W) using its internal histogram engine. Then, the built-in algorithm extracts the flight time corresponding to the peak values, which can be converted into the distance value (in mm) for that grid. Next, the data from all grids is arranged in spatial order to form the input for each frame. .

[0042] The generated feature maps can then be processed by global average pooling and can be projected through a fully connected layer to generate compact frame-level descriptors.

[0043]

[0044] After aggregating these descriptors over time, a sequence representation of size (T, D) is obtained, where T represents the length of the input frames fed into the temporal branch, also known as the number of input frames, and T can be, but is not limited to, 200; D is the dimension of the frame-level descriptors obtained after extraction and compression by the network, and D can be, but is not limited to, 512. This sequence can then be processed by a bidirectional long short-term memory network encoder to capture long-term temporal dependencies. In this disclosure, instead of relying solely on the final hidden state, a temporal attention mechanism can be introduced to adaptively aggregate all hidden states through learnable attention weights. This enables the model to highlight information-rich frames and enhance its ability to represent significant dynamic patterns.

[0045]

[0046] , In the formula, This represents the hidden state of a bidirectional long short-term memory network. Represents the memory unit of a bidirectional long short-term memory network. Indicates the attention coefficient. This represents the weighted global temporal characteristics.

[0047] In step 404, the reflected pulse signal can be input into the time-frequency domain branch to obtain time-frequency domain features, wherein the time-frequency domain features are used to characterize the frequency variation of the reflected pulse signal over time.

[0048] According to exemplary embodiments of this disclosure, as previously described, the time-frequency domain branch may include a short-time Fourier transform module, a convolutional neural network, an adaptive average pooling module, and a fully connected layer.

[0049] First, the reflected pulse signal can be input into a short-time Fourier transform module to obtain a time-frequency plot. Then, the time-frequency plot can be input into a convolutional neural network to obtain local time-frequency features. Next, the local time-frequency features can be input into an adaptive average pooling module to obtain adaptive average pooling features. Finally, the adaptive average pooling features can be input into a fully connected layer to obtain time-frequency domain features.

[0050] In the time-frequency domain branch, for each pixel (i, j) of the multi-zone ToF sensor, a short-time Fourier transform can be applied to its distance sequence of length T to generate a spectrum. Then, the complex spectrum can be converted into a modulus representation, and further transformed using logarithmic compression to reduce the dynamic range and enhance weak frequency components.

[0051]

[0052]

[0053]

[0054] In the formula, For a window function (e.g., the Hanning window), t is the time-shift index, and f represents the frequency variable corresponding to the discrete spectral unit in the FFT-based computation.

[0055] The next step is the time-frequency domain feature extraction stage. First, the time-frequency map of each pixel can be extracted. It is considered a single-channel input and can be organized into a shape of (H) A tensor of W,1,f,t), where The tensors of each pixel channel can be used to extract local time-frequency features (H) through a convolutional neural network. The features are arranged in a matrix (W, C), where C represents the number of channels in the convolutional neural network. For example, C can be, but is not limited to, 32. Then, the features can be rearranged according to the sensor's geometric layout to form a two-dimensional feature map of shape (C, H, W), thus preserving spatial structure. Next, adaptive average pooling can be applied to reduce the spatial resolution from 8×8 to 4×4 to generate a compact and spatially aggregated representation. Then, the features are pooled... After flattening, the final representation can be obtained through a fully connected layer.

[0056] In step 405, the time-domain features and time-frequency domain features can be fused to obtain fused features.

[0057] To model the dependency between time-domain and time-frequency domain representations, this disclosure employs a lightweight cross-modal gating mechanism. First, time-domain features can be projected onto the time-frequency feature space and concatenated with them. Then, weight coefficients can be predicted using a feedforward network. This is used to adaptively adjust the weights of the time-frequency domain features. Next, a fully connected fusion network can be used to fuse the gated time-frequency domain features with the original time-domain features:

[0058]

[0059] In the formula, This represents a lightweight fusion network that can consist of two fully connected layers and can employ the ReLU activation function and dropout mechanism. Represents the learnable network parameters of a multilayer perceptron (MLP). This represents the aforementioned weighted global temporal characteristics. This indicates the aforementioned pooling characteristics, i.e., frequency domain characteristics.

[0060] In step 406, the adsorption properties of the target to be adsorbed can be evaluated based on the fusion characteristics.

[0061] Compared to single-modal sensing methods, the fusion sensing method proposed in this disclosure can achieve stable vibration estimation under various motion modes and different sensing conditions, thus providing a solid theoretical foundation and reliable technical guidance for near-field target perception of underwater robots.

[0062] This disclosure also allows for multi-task joint optimization, meaning that it can jointly model the geometric surface properties of the adsorbed object and its vibration dynamics. Specifically, surface type and vibration mode can be considered as classification tasks in discrete state space, while vibration frequency estimation can be modeled as a regression task in a continuous domain. Accordingly, the perceptual framework can be constructed as a multi-task learning problem with both regression and classification objectives. The aforementioned fused feature vector... It can be input to four independent fully connected heads: three main heads for surface type classification, vibration type classification, and vibration frequency regression, respectively; and one auxiliary head for discrete classification of vibration frequency. The auxiliary classification branch is introduced to enhance the spectral feature representation.

[0063] In addition, all classification heads can be optimized using the standard cross-entropy loss function:

[0064] In the formula, and Let N represent the true label and predicted probability of class c in sample i, respectively. C is the total number of classes for each task, and N is the number of samples. This represents the classification task loss of the target surface to be adsorbed. The classification task loss represents the vibration type of the target to be adsorbed. The classification task loss represents the vibration frequency of the target to be adsorbed.

[0065] Furthermore, the loss in the vibration frequency regression task can be represented by the mean-square error (MSE):

[0066] In the formula, and These represent the actual and predicted values ​​of the vibration frequency, respectively. For the sample size, This represents the regression task loss of the vibration frequency.

[0067] The final joint optimization objective integrates the four loss terms mentioned above to form a multi-task framework, thereby enhancing the model's ability to capture both geometric and dynamic representations simultaneously.

[0068] In the formula, ~ These are task-specific weighting coefficients, primarily used to balance the contributions of each subtask.

[0069] According to exemplary embodiments of this disclosure, the adsorption properties may include the surface geometry of the target to be adsorbed and / or the motion characteristics of the target to be adsorbed. For example, this disclosure primarily focuses on the motion patterns that the adsorbed object may exhibit in an underwater environment, such as pitching, lateral movement, and periodic swaying.

[0070] According to exemplary embodiments of this disclosure, in order to achieve robust sensing and accurate identification of the dynamic surface of the target to be adsorbed, the surface of the target to be adsorbed can be classified into the following types: planar and curved.

[0071] According to exemplary embodiments of this disclosure, motion characteristics may include vibration type and / or vibration frequency, and vibration type may be divided into translational vibration and rotational vibration. That is, in this disclosure, in order to achieve robust sensing and accurate discrimination of dynamic surfaces, their rhythmic motion can be summarized into two basic forms: translational vibration and rotational vibration.

[0072] This disclosure provides a biomimetic robotic fish near-field perception method based on multi-region Time-of-Flight (ToF) sequence information, aiming to solve the problem of low visual reliability in near-field perception scenarios of robotic fish in related technologies. The core of this disclosure lies in a spatiotemporal perception model based on multi-region ToF sequence information, which can simultaneously capture temporal and time-frequency characteristics, thereby enabling accurate estimation of typical surface features and motion patterns in near-field interactions. Specifically, the temporal branch can extract spatial features through residual CNN, capture temporal dependencies through BiLSTM, and enhance keyframe features by combining temporal attention; the time-frequency branch can generate a time-spectrum map through STFT, extract features through CNN, and undergo reconstruction processing to obtain global spectral features. Finally, the temporal and time-frequency feature layers can be fused to achieve complementary integration, thereby effectively supporting the joint prediction of attributes such as typical surface curvature and typical motion patterns of the target.

[0073] It should be noted that, in this disclosure, to alleviate the difficulty of acquiring underwater ToF sequences, near-field sensing data of typical adsorbed objects can be further constructed in the simulation. Specifically, based on the evaluation of the noise characteristics of the real VL53L7CX sensor, a Gaussian noise model can be incorporated into the simulator to ensure that the generated multi-region time-of-flight series exhibits variance and perturbation patterns consistent with real underwater sensing. Furthermore, the simulation operating range can follow the sensor's effective underwater near-field region (20-60 mm), within which the measurement standard deviation is approximately 0.4 mm. Table 1 summarizes six simulation tasks, covering geometric attribute identification and motion parameter estimation. The target motion pattern can include axial reciprocating translation and rotational vibration, and its joint modeling expression can be:

[0074] In the formula, Represents linear reciprocating translation or around the axis rotational component Furthermore, the rotational component can be further converted into a quaternion form for attitude control, and the vibration frequency f can cover the range of 0-2 Hz with a step size of 0.1 Hz.

[0075] Table 1 Simulated motion of the adsorbed objects

[0076] Figure 6 This is a schematic diagram comparing the estimated distributions of a single-modal (time domain, time-frequency domain) sensing model in related technologies with the fusion spatiotemporal sensing model provided in this disclosure for six typical motion types. (Refer to...) Figure 6 Each subplot can be a two-dimensional histogram showing the predicted frequency versus the actual frequency, with brighter areas representing higher occurrence densities. Figure 6 As can be seen, the predicted values ​​are closely distributed along the diagonal in all categories, which indicates that the root mean square error is consistently low, thus confirming that the model has high accuracy in frequency estimation.

[0077] Table 2. Root mean square error / variance results for different models in six types of data.

[0078] Table 3 Experimental results of frequency estimation under noise and distribution offset conditions

[0079] Table 2 shows the root mean square errors (RMSEs) of the single-modal (time domain and time-frequency domain) and the proposed fusion spatiotemporal perception model on the original validation set. As shown in Table 2, the single time-domain model exhibits the smallest variance, indicating stable predictive ability across different samples; the single time-frequency model shows slightly larger dispersion, reflecting its increased sensitivity to time window effects; the proposed fusion spatiotemporal perception model achieves a balanced trade-off, with its RMS and standard deviation falling between those of the two single-domain benchmark models.

[0080] Table 3 adds additive noise and systematic bias to the original validation set. Additive noise introduces random perturbations, while bias simulates distribution shifts. Table 3 shows that the single time-frequency model maintains a relatively low root mean square error as noise levels gradually increase, reflecting the robustness of the spectral representation to transient disturbances. However, its performance significantly degrades under systematic bias conditions (Bias-0.3 and Bias-0.5), indicating that persistent baseline shifts distort the underlying spectral energy distribution. The single time model exhibits increased variance under both noise and bias conditions, indicating its sensitivity to non-stationary perturbations. In contrast, the fused spatiotemporal sensing model proposed in this disclosure achieves stable performance under all settings by combining temporal continuity and spectral invariance. Specifically, the time-spectral branch mitigates random noise, while the time-domain branch preserves local dynamic patterns, thereby enhancing robustness to both additive and distributed noise.

[0081] To quantify stability, this disclosure also introduces a standardized metric, Score. minmax The metric can be calculated by performing minimum-maximum normalization of the root mean square error (RMSE) on the model at each perturbation level, and by taking the average of the normalization scores under all conditions. This metric is mainly used to reflect the average relative robustness of the model under different perturbation intensities. This metric demonstrates that the model proposed in this disclosure achieves a more balanced trade-off between accuracy and robustness.

[0082] To further verify the practical application effect of the fusion spatiotemporal perception model proposed in this disclosure, a verification experiment was also conducted in a real underwater scenario. The relevant experimental details and results are as follows: Figure 7 As shown. Figure 7 This is a graph illustrating the frequency estimation results of a spatiotemporal perception model according to an exemplary embodiment of the present disclosure in a real-world scenario under different target distances and dynamic frequency variations. (Refer to...) Figure 7The raw experimental data were collected at a sampling rate of 15 Hz, consistent with the simulation experiment settings. Measurement data were collected at four representative distances: 25 mm, 30 mm, 40 mm, and 60 mm, comprehensively covering the main working range in the simulation experiment. Furthermore, to ensure the statistical reliability of the experimental data, ten independent motion sequences were recorded for each distance condition. Experimental results show that the vibration frequency predicted by the model is close to the actual vibration frequency under all test conditions, with an average root mean square error of only 0.20 Hz.

[0083] In addition, refer to Figure 7 The experiment also presented the model's performance within a frequency range of 0.68 Hz to 1.25 Hz. The estimated values ​​were stepped every 15 frames, with frequency tracking results ranging from 0.84 Hz to 1.12 Hz. These results confirm the effectiveness of the spatiotemporal perception model proposed in this disclosure in real underwater scenarios, highlighting its potential for near-field perception in robotic adsorption tasks.

[0084] Figure 8 This is a block diagram illustrating an evaluation device 800 for adsorption properties according to an exemplary embodiment of the present disclosure, applied to a biomimetic underwater robotic fish. The biomimetic underwater robotic fish may be equipped with a multi-zone ToF sensor and a spatiotemporal sensing model, which may include a time-domain branch and a time-frequency domain branch.

[0085] Reference Figure 8 The evaluation device 800 may include a signal transmission module 801, a signal receiving module 802, a time-domain feature acquisition module 803, a time-frequency domain feature acquisition module 804, a fusion module 805, and a feature evaluation module 806.

[0086] The signal transmission module 801 can transmit pulse signals to the target to be adsorbed through a multi-zone ToF sensor.

[0087] The signal receiving module 802 can receive the reflected pulse signal obtained by the pulse signal reflected by the target to be adsorbed through the multi-zone ToF sensor.

[0088] The time-domain feature acquisition module 803 can input the reflected pulse signal into the time-domain branch to obtain time-domain features, which can be used to characterize the variation of the amplitude of the reflected pulse signal over time.

[0089] According to exemplary embodiments of this disclosure, as previously described, the temporal branch may include a two-layer residual convolutional block, a global average pooling module, a fully connected module, and a bidirectional long short-term memory network.

[0090] First, the temporal feature acquisition module 803 can input the reflected pulse signal into a two-layer residual convolutional block to obtain local spatial features. Then, the temporal feature acquisition module 803 can input the local spatial features into a global average pooling module to obtain global average pooling features. Next, the temporal feature acquisition module 803 can input the global average pooling features into a fully connected module to obtain fully connected features. Finally, the temporal feature acquisition module 803 can input the fully connected features into a bidirectional long short-term memory network to capture long-term temporal dependencies as temporal features.

[0091] The time-frequency domain feature acquisition module 804 can input the reflected pulse signal into the time-frequency domain branch to obtain time-frequency domain features, which are used to characterize the frequency variation of the reflected pulse signal over time.

[0092] According to exemplary embodiments of this disclosure, as previously described, the time-frequency domain branch may include a short-time Fourier transform module, a convolutional neural network, an adaptive average pooling module, and a fully connected layer.

[0093] First, the time-frequency domain feature acquisition module 804 can input the reflected pulse signal into the short-time Fourier transform module to obtain a time-frequency graph. Then, the time-frequency domain feature acquisition module 804 can input the time-frequency graph into a convolutional neural network to obtain local time-frequency features. Next, the time-frequency domain feature acquisition module 804 can input the local time-frequency features into an adaptive average pooling module to obtain adaptive average pooling features. Finally, the time-frequency domain feature acquisition module 804 can input the adaptive average pooling features into a fully connected layer to obtain time-frequency domain features.

[0094] The fusion module 805 can fuse time-domain features and time-frequency domain features to obtain fused features.

[0095] The characteristic evaluation module 806 can evaluate the adsorption properties of the target to be adsorbed based on the fusion characteristics.

[0096] According to exemplary embodiments of this disclosure, the adsorption properties may include the surface geometry of the target to be adsorbed and / or the motion characteristics of the target to be adsorbed. For example, this disclosure primarily focuses on the motion patterns that the adsorbed object may exhibit in an underwater environment, such as pitching, lateral movement, and periodic swaying.

[0097] According to exemplary embodiments of this disclosure, in order to achieve robust sensing and accurate identification of the dynamic surface of the target to be adsorbed, the surface of the target to be adsorbed can be classified into the following types: planar and curved.

[0098] According to exemplary embodiments of this disclosure, motion characteristics may include vibration type and / or vibration frequency, and vibration type may be divided into translational vibration and rotational vibration. That is, in this disclosure, in order to achieve robust sensing and accurate discrimination of dynamic surfaces, their rhythmic motion can be summarized into two basic forms: translational vibration and rotational vibration.

[0099] Figure 9 This is a block diagram illustrating an electronic device 900 according to an exemplary embodiment of the present disclosure.

[0100] Reference Figure 9 The electronic device 900 includes at least one memory 901 and at least one processor 902. The at least one memory 901 stores instructions that, when executed by the at least one processor 902, perform an evaluation method for adsorption properties according to an exemplary embodiment of the present disclosure.

[0101] As an example, electronic device 900 can be a PC, tablet, personal digital assistant, smartphone, or other device capable of executing the aforementioned instructions. Here, electronic device 900 is not necessarily a single electronic device, but can be any collection of devices or circuits capable of executing the aforementioned instructions (or instruction sets) individually or in combination. Electronic device 900 can also be part of an integrated control system or system manager, or can be configured to interconnect with a portable electronic device locally or remotely (e.g., via wireless transmission) through an interface.

[0102] In electronic device 900, processor 902 may include a central processing unit (CPU), a graphics processing unit (GPU), a programmable logic device, a dedicated processor system, a microcontroller, or a microprocessor. By way of example and not limitation, processor may also include analog processors, digital processors, microprocessors, multi-core processors, processor arrays, network processors, etc.

[0103] The processor 902 can execute instructions or code stored in the memory 901, which can also store data. Instructions and data can also be sent and received via a network through a network interface device, which can employ any known transmission protocol.

[0104] The memory 901 may be integrated with the processor 902, for example, by placing RAM or flash memory within an integrated circuit microprocessor. Alternatively, the memory 901 may include a separate device, such as an external disk drive, a storage array, or other storage device usable by any database system. The memory 901 and the processor 902 may be operatively coupled, or may communicate with each other, for example, via I / O ports, network connections, etc., enabling the processor 902 to read files stored in the memory.

[0105] In addition, the electronic device 900 may also include a video display (such as a liquid crystal display) and a user interaction interface (such as a keyboard, mouse, touch input device, etc.). All components of the electronic device 900 can be interconnected via a bus and / or network.

[0106] According to exemplary embodiments of this disclosure, a computer-readable storage medium may also be provided, which, when executed by a processor of an electronic device, enables the electronic device to perform the aforementioned method for evaluating the absorbability characteristics. Examples of computer-readable storage media include: read-only memory (ROM), random access programmable read-only memory (PROM), electrically erasable programmable read-only memory (EEPROM), random access memory (RAM), dynamic random access memory (DRAM), static random access memory (SRAM), flash memory, non-volatile memory, CD-ROM, CD-R, CD+R, CD-RW, CD+RW, DVD-ROM, DVD-R, DVD+R, DVD-RW, DVD+RW, DVD-RAM, BD-ROM, BD-R, BD-R LTH, BD-RE, Blu-ray or optical disc storage, hard disk drive (HDD), solid-state drive (SSD), card storage (such as multimedia cards, secure digital (SD) cards, or ultra-fast digital (XD) cards), magnetic tape, floppy disk, magneto-optical data storage device, optical data storage device, hard disk, solid-state drive, and any other device configured to store a computer program and any associated data, data files, and data structures in a non-transitory manner and to provide the computer program and any associated data, data files, and data structures to a processor or computer so that the processor or computer can execute the computer program. The computer program in the aforementioned computer-readable storage medium can run in an environment deployed in computer devices such as clients, hosts, agent devices, servers, etc. Furthermore, in one example, the computer program and any associated data, data files, and data structures are distributed across a networked computer system, such that the computer program and any associated data, data files, and data structures are stored, accessed, and executed in a distributed manner through one or more processors or computers.

[0107] According to exemplary embodiments of the present disclosure, a computer program product may also be provided, including a computer program that, when executed by a processor, implements a method for evaluating the adsorption properties according to the present disclosure.

[0108] According to the method, apparatus, electronic device, storage medium, and computer program product for evaluating adsorption properties disclosed herein, by setting a multi-zone ToF sensor as the core sensing unit on a biomimetic underwater robotic fish, the reflected pulse signals reflected back by the target to be adsorbed can be continuously acquired, thereby constructing a multi-dimensional input that combines spatial geometric information and temporal dynamic characteristics. Furthermore, by collaboratively modeling this multi-dimensional input through parallel time-domain and time-frequency domain branches included in the spatiotemporal sensing model, complementary characterization of temporal evolution patterns and spectral characteristics can be achieved, thereby enabling joint sensing of the adsorption properties of the target and improving the accuracy of the evaluation of the host's adsorption properties.

[0109] According to exemplary embodiments of the present disclosure, compared with single-modal sensing methods, the fusion sensing method proposed in the present disclosure can achieve stable vibration estimation under multiple motion modes and different sensing conditions, thereby providing a solid theoretical foundation and reliable technical guidance for near-field target perception of underwater robots.

[0110] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This disclosure is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the following claims.

[0111] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is limited only by the appended claims.

Claims

1. A method for evaluating adsorption properties, applied to a biomimetic underwater robotic fish, characterized in that, The biomimetic underwater robotic fish is equipped with a multi-zone ToF sensor and a spatiotemporal sensing model. The spatiotemporal sensing model includes a time-domain branch and a time-frequency domain branch. The evaluation method includes: The multi-zone ToF sensor emits pulse signals to the target to be adsorbed. The multi-zone ToF sensor receives the pulse signal and obtains the reflected pulse signal by the target to be adsorbed; The reflected pulse signal is input into the time domain branch to obtain time domain features, wherein the time domain features are used to characterize the variation of the amplitude of the reflected pulse signal with time; The reflected pulse signal is input into the time-frequency domain branch to obtain time-frequency domain features, wherein the time-frequency domain features are used to characterize the frequency variation of the reflected pulse signal over time. The time-domain features and the time-frequency domain features are fused to obtain fused features; Based on the fusion characteristics, the adsorption properties of the target to be adsorbed are evaluated.

2. The evaluation method as described in claim 1, characterized in that, The temporal branch includes a two-layer residual convolutional block, a global average pooling module, a fully connected module, and a bidirectional long short-term memory network; The step of inputting the reflected pulse signal into the time domain branch to obtain time domain characteristics includes: The reflected pulse signal is input into the two-layer residual convolutional block to obtain local spatial features; The local spatial features are input into the global average pooling module to obtain global average pooling features; The global average pooling features are input into the fully connected module to obtain the fully connected features; The fully connected features are input into the bidirectional long short-term memory network to capture long-term temporal dependencies as the temporal features.

3. The evaluation method as described in claim 1, characterized in that, The time-frequency domain branch includes a short-time Fourier transform module, a convolutional neural network, an adaptive average pooling module, and a fully connected layer; The step of inputting the reflected pulse signal into the time-frequency domain branch to obtain time-frequency domain features includes: The reflected pulse signal is input into the short-time Fourier transform module to obtain a time-frequency diagram; The time-frequency graph is input into the convolutional neural network to obtain local time-frequency features; The local time-frequency features are input into the adaptive average pooling module to obtain adaptive average pooling features; The adaptive average pooling features are input into the fully connected layer to obtain the time-frequency domain features.

4. The evaluation method as described in claim 1, characterized in that, The adsorption properties include the surface geometry of the target to be adsorbed and / or the motion characteristics of the target to be adsorbed.

5. The evaluation method as described in claim 4, characterized in that, The surface geometric properties include the following types: Plane, curved surface.

6. The evaluation method as described in claim 4, characterized in that, The motion characteristics include vibration type and / or vibration frequency, and the vibration type is divided into translational vibration and rotational vibration.

7. An evaluation device for adsorption properties, applied to a biomimetic underwater robotic fish, characterized in that, The biomimetic underwater robotic fish is equipped with a multi-zone ToF sensor and a spatiotemporal sensing model, the spatiotemporal sensing model including a time-domain branch and a time-frequency domain branch, and the evaluation device includes: The signal transmission module is configured to transmit pulse signals to the target to be adsorbed via the multi-zone ToF sensor; The signal receiving module is configured to receive the reflected pulse signal obtained by the pulse signal being reflected by the target to be adsorbed through the multi-zone ToF sensor; The time-domain feature acquisition module is configured to input the reflected pulse signal into the time-domain branch to obtain time-domain features, wherein the time-domain features are used to characterize the variation law of the amplitude of the reflected pulse signal over time; The time-frequency domain feature acquisition module is configured to input the reflected pulse signal into the time-frequency domain branch to obtain time-frequency domain features, wherein the time-frequency domain features are used to characterize the frequency variation law of the reflected pulse signal over time. The fusion module is configured to fuse the time-domain features and the time-frequency domain features to obtain fused features; The characteristic evaluation module is configured to evaluate the adsorption properties of the target to be adsorbed based on the fusion characteristics.

8. An electronic device, characterized in that, include: processor; Memory used to store the processor's executable instructions; The processor is configured to execute the instructions to implement the method for evaluating the adsorption properties as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, When the instructions in the computer-readable storage medium are executed by the processor of the electronic device, the electronic device is able to perform the method for evaluating the adsorption properties as described in any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method for evaluating the adsorption properties as described in any one of claims 1 to 6.

Images