Tracked robot intelligent reconnaissance system based on panoramic vision and multispectral analysis

Abstract

The invention relates to the technical field of robot reconnaissance, and discloses a tracked robot intelligent reconnaissance system based on panoramic vision and multispectral analysis, which comprises the following steps: dynamically adjusting sensor parameters based on environmental spectral characteristics to adapt to different light conditions, and collecting high-quality panoramic vision and multispectral data; performing panoramic vision fusion processing on the acquired data to compensate light change in a complex environment and enhance image quality; performing multispectral data analysis according to the fused data, and classifying target materials based on an analysis result to improve reconnaissance accuracy; and a relevance optimization algorithm of visual data and spectral information is combined to filter interference in multi-target reconnaissance and feed back and correct path planning of the robot in real time. Target materials are classified according to a multispectral data analysis result to solve the problem of insufficient target recognition accuracy in the reconnaissance process, and the success rate and high efficiency of a reconnaissance task are ensured by combining machine learning with a prediction algorithm.

Description

Technical Field

[0001] The present invention relates to the technical field of robot reconnaissance, and particularly to a tracked robot intelligent reconnaissance system based on panoramic vision fusion and multispectral analysis. Background Art

[0002] The tracked robot intelligent reconnaissance system based on panoramic vision fusion and multispectral analysis is a technical means that utilizes panoramic vision and multispectral analysis technologies, combined with the mobility flexibility of the tracked robot, to achieve intelligent reconnaissance of complex environments.

[0003] However, this method still faces some challenges. First, it is how to accurately classify target materials through multispectral data analysis to improve the accuracy of target recognition during the reconnaissance process. Second, the ability to compensate for light changes in complex environments is insufficient, which will affect the stability of image quality. Third, under specific conditions, due to the failure to dynamically adjust sensor parameters to match the environmental spectral characteristics, the reconnaissance accuracy decreases.

[0004] In addition, when facing simultaneous reconnaissance of multiple targets, due to the lack of efficient algorithm optimization, the correlation between visual data and spectral information is insufficient and it is difficult to filter out interference. Finally, the coordination between the real-time feedback of reconnaissance data and the robot path planning is not tight enough, which may bring the risk of target loss. Solving these problems will be the key to improving this intelligent reconnaissance method. Summary of the Invention

[0005] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0006] A tracked robot intelligent reconnaissance system based on panoramic vision fusion and multispectral analysis, comprising:

[0007] Dynamically adjusting sensor parameters based on environmental spectral characteristics to adapt to different light conditions and collect high-quality panoramic vision and multispectral data; performing panoramic vision fusion processing on the collected data to compensate for light changes in complex environments and enhance image quality; performing multispectral data analysis based on the fusion data, classifying target materials based on the analysis results to improve reconnaissance accuracy; optimizing the algorithm by combining the correlation between visual data and spectral information to filter out interference in multi-target reconnaissance and real-time feedback to correct the robot's path planning;

[0008] The dynamic adjustment of sensor parameters based on environmental spectral characteristics further includes: determining the main frequency component f_main of the current environmental spectrum; calculating the spectral distribution variance σ2 = 1 / N∑(f_i - f_mean)2, where N is the number of frequency points, f_i represents the intensity of the i-th frequency point, and f_mean is the average frequency intensity; when σ2 > threshold1, adjusting the sensor sensitivity to k×σ2, where k is the sensitivity coefficient; calibrating the sensor to ensure data consistency and accuracy.

[0009] Preferably, the panoramic vision fusion processing of the collected data includes:

[0010] According to the image pixel distribution, calculating the scene illumination deviation ΔL = |I_max - I_min| / I_ave, where I_max and I_min are the maximum and minimum brightness values respectively, and I_ave is the average brightness value;

[0011] If ΔL exceeds the threshold threshold2, then use the global brightness compensation formula C_compensate = C_original + λ×ΔL for image enhancement, where λ is the compensation factor;

[0012] Applying a panoramic stitching algorithm to seamlessly stitch the images into a 360° view;

[0013] Using edge-preserving filtering to eliminate artifacts and enhance the detail quality in complex environments.

[0014] Preferably, the multispectral data analysis based on the fusion data includes:

[0015] Using a normalization algorithm to process the energy values E_i of different bands to obtain E_norm(i) = (E_i - E_min) / (E_max - E_min);

[0016] Defining the spectral matching degree D of the target substance as D = Σ|S_ref(i) - E_norm(i)|, where S_ref(i) is the spectral data of the reference material;

[0017] Selecting the category of S_ref(i) corresponding to the minimum D as the material type;

[0018] Adding the material type to a probability classification module to adapt to uncertainty.

[0019] Preferably, the correlation optimization algorithm combining visual data and spectral information includes:

[0020] Constructing a joint feature space J = [V1, V2,..., VS], where Vi represents the position of each target in the feature vector space;

[0021] The weight \(w_j=\sum cov(F_{vis}(j),F_{spec}(j)) / \sum var(F_{vis}(j))\) is obtained through covariance analysis, where \(F_{vis}\) and \(F_{spec}\) represent the variances of the visual and spectral feature dimensions respectively;

[0022] Set the interference rejection threshold \(threshold4\) and reject interference through the following condition: \(w_j < threshold4\);

[0023] Retrain the classifier based on the remaining high-confidence features for accurate target recognition.

[0024] Preferably, the interference filtering in multi-target reconnaissance further includes:

[0025] Map the interference intensity signal collected in real time to the interference level index \(M\), where \(M = \log(\sum|I_{noise}I_{baseline}|^2) / \log(N_{noise})\), and \(N_{noise}\) represents the total number of interference frames;

[0026] When \(M > threshold5\), start the redundant data verification logic to remove invalid frames;

[0027] Extract the relative motion feature \(H_{diff}(j)\) between multiple targets in the valid frames to distinguish stationary and moving entities;

[0028] Prioritize high-value moving targets according to the classification results to optimize the reconnaissance efficiency.

[0029] Preferably, the optimized interference conditions are:

[0030] Introduce a self-learning adjustment module to update the evaluation formula of the interference level as \(A=\alpha M+\beta\sum H_{rel}\), where \(\alpha\) and \(\beta\) are weight parameters, and \(H_{rel}\) is the sum of relative relationship features;

[0031] If \(A > threshold6\), enable the backup sensing system to ensure reliability;

[0032] Integrate redundant sensing data into the existing reconnaissance model to form a closed-loop correction structure;

[0033] Improve the target localization resolution \(R_{loc}\geq resolution_{goal}\) in the key frame detection stage.

[0034] Preferably, the real-time feedback correction path planning includes:

[0035] Define the proportional deviation of the current position from the reference trajectory as \(P_{deviation}=|pos_{current}-pos_{benchmark}| / path_{len_{total}}\);

[0036] When P_deviation > threshold7 or the obstacle distance D_obst ≤ threshold8, start the replanning program;

[0037] Generate multiple candidate routes r_candidate and filter them according to the estimated cost function COST(r_candidate, t_now);

[0038] Select the path with the lowest cost and update the historical data to support the improvement of prediction ability.

[0039] Preferably, the intelligent correction step of the enhanced path is as follows:

[0040] Establish a prediction window T_future = time_remain * γ (the time window magnification γ is set according to the task priority) to predict the set of possible problem points P_future in the future;

[0041] If any p ∈ P_future satisfies the conflict judgment formula C(p_i, t_estimtated, v_target) > critical_conflict_value, avoid it in advance;

[0042] Introduce a two-level obstacle avoidance rule to first locally adjust the angle and then globally adjust the direction to avoid excessive detours;

[0043] After the adjustment is completed, perform the safety detection again until the standard condition exit_flag = true is satisfied.

[0044] Preferably, improving the multi-camera synchronization efficiency includes:

[0045] Set the camera synchronization signal trigger delay τ_synchronization and control the actual delay range stability by the formula τ_sync_actual = ceil[Δτ / step_unit)] (step_unit is the unit step size);

[0046] When the frame rate change exceeds the allowable range δfps = |fps_current - fps_designed| / fps_designed > threshold_fps, trigger the warning mechanism and adjust the light source intensity;

[0047] Ensure that the imaging quality balance error err_quality of all channels does not exceed the set tolerance interval_tolerance;

[0048] Achieve fast response while reducing the energy consumption to meet the requirements of the long-term operating environment.

[0049] Preferably, the steps for enhancing the repair of image quality problems include:

[0050] Record the global exposure parameter exp_global and the local brightness ratio factor_bright_local = avg_intensity_high / avg_intensity_low;

[0051] Apply the histogram correction formula hist_new(z) = exp_global × factor_correction × factor_bright_local to the high-contrast region;

[0052] Integrate the deep learning model to fine-tune the exposure and sharpening effects and output the optimized image Q_optimize = net(Q_initial, fine_tuning_params);

[0053] Regularly loop back to verify the effectiveness of the repair strategy to ensure long-term consistent stability performance.

[0054] Compared with the prior art, the beneficial effects achieved by the present invention are:

[0055] 1. Dynamically adjust the sensor parameters based on the environmental spectral characteristics to adapt to different light conditions and collect high-quality panoramic vision and multispectral data; perform panoramic vision fusion processing on the collected data to compensate for the light changes in complex environments and enhance the image quality; perform multispectral data analysis based on the fusion data, classify the target materials based on the analysis results to improve the reconnaissance accuracy; optimize the algorithm by combining the correlation between visual data and spectral information to filter out the interference in multi-target reconnaissance and provide real-time feedback to correct the path planning of the robot.

[0056] 2. By combining machine learning and prediction algorithms, analyze the current target reconnaissance data and historical path records, and dynamically correct the robot path. This enables the tracked robot to still be able to track and avoid obstacles or misjudgment points in a timely manner even when the target position changes, reducing the risk of target loss, thereby ensuring the success rate and efficiency of the reconnaissance mission. In summary, this reconnaissance method comprehensively improves the adaptability and reconnaissance ability of the tracked robot from multiple dimensions. Brief Description of the Drawings

[0057] Figure 1 It is a schematic flowchart of the reconnaissance system of the present invention;

[0058] Figure 2 It is a schematic flowchart of the target recognition process of the present invention;

[0059] Figure 3 It is; a schematic flowchart of the path data process of the present invention. Detailed Embodiments

[0060] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0061] Please refer to Figure 1 - Figure 3 , for the intelligent reconnaissance system of the tracked robot based on panoramic vision fusion and multispectral analysis of the present invention, first, it is necessary to dynamically adjust the sensor parameters to collect high-quality data, then perform fusion processing on the images under complex lighting conditions, and analyze the target material based on multispectral data. At the same time, combine the correlation optimization algorithm to filter out interference, and finally correct the path planning according to the real-time feedback to improve the reliability.

[0062] First, dynamically adjusting the sensor parameters based on the environmental spectral characteristics is one of the key steps to ensure the accuracy and reliability of the data. During the specific operation process, the multispectral sensor installed at the front end of the robot collects the spectral characteristic information of a specific area environment and analyzes the light source wavelength and intensity distribution. This step can sense the environmental brightness change and the existence of special light sources (such as ultraviolet or infrared radiation). Then, the system automatically corrects the working parameters of the camera, lidar or other sensing modules, such as exposure time, gain value and filter settings, according to the conclusion of the spectral analysis. This method enables the tracked robot to cope with diverse and complex external environmental conditions and maintain a high level of data collection quality.

[0063] Second, perform panoramic vision fusion processing on the collected data to compensate for the image degradation problem caused by uneven light. Match multiple groups of raw data from different perspectives or different bands (such as visible light and infrared) in a unified coordinate system, and reduce the distortion error or noise accumulation phenomenon brought by the overlapping area between each other through algorithms. At the same time, use advanced contrast stretching technology and local mean balance technology to further enhance the light and dark level performance of the picture, achieving an ideal detail presentation effect, thereby effectively improving the visibility and discrimination ability in the overall reconnaissance task.

[0064] To achieve target material classification to improve the reconnaissance accuracy, it is also necessary to carry out meticulous multispectral data analysis on the basis of the fused image. It is mainly to identify the pattern differences in the reflection characteristics of each potential object in each band through a model trained by machine learning, and then use statistical laws to infer the possible corresponding material types, such as the category attribution relationship of metal products or natural rocks. In one embodiment, assuming that an unknown object appears in the reconnaissance site, the corresponding spectral scan results can be retrieved and compared with the standard templates stored in the database in advance to evaluate and confirm the attribute category, significantly reducing the false alarm rate and improving the judgment accuracy at the same time.

[0065] It is also very necessary to optimize the algorithm design by combining the correlation characteristics between visual data and spectral information. This is especially true in challenging situations where multiple targets exist simultaneously. For this purpose, a specially customized weight factor formula is introduced to comprehensively consider multiple aspects such as the proximity of spatial positions, the similarity level of shapes, and the matching probability of reflection characteristics, and to construct a joint evaluation basis for screening and eliminating false prompt information, only retaining the most relevant part as the basis for further processing. Specifically, when faced with the situation of various shell fragments scattered disorderly on the battlefield, relying solely on ordinary shape recognition will inevitably lead to omissions. However, by additionally considering approximate absorption and release energy characteristics, the key elements of real interest can be successfully distinguished, forming a more precise focusing direction to guide the next action decision towards a mature, stable, accurate, and reliable development trend.

[0066] The last step, which is also one of the core elements indispensable in the entire reconnaissance process, is to adjust the current travel trajectory in a timely manner based on the continuously generated real-time reconnaissance data, so as to avoid falling into a passive situation again and causing the risk of losing important clues outside the locked range. This flexible response mechanism requires the equipped with an efficient communication unit to ensure unobstructed two-way interaction between the inside and outside. Coupled with a pre-set reserve of multiple emergency plan options, it supports switching to execute the optimal action strategy according to the actual situation, and the operation process is stable and smooth. For example, when detecting that the line of sight is suddenly blocked by dense obstacles ahead, the clarity is severely reduced, and the success rate of completing the expected mission is seriously threatened, the built-in obstacle avoidance planning subsystem is quickly activated to regenerate a detour alternative path selection method to ensure that the final destination can still be reached according to the requirements of the initially established task framework, complete all the scheduled inspection content coverage requirements, and successfully conclude the entire process, achieving a complete closed-loop management that is standardized, scientific, advanced, reliable, efficient, and with remarkable results.

[0067] The system for dynamically adjusting sensor parameters based on environmental spectral characteristics of the present invention specifically includes four steps, and the functions of each step are elaborated from two aspects: formula calculation and practical application.

[0068] The first step is to determine the main frequency component of the current environmental spectrum, analyze the frequency point where the intensity of the main light source existing in the current environment is located, and use this frequency point as the key processing target. In a multi-light source scenario, such as an urban area or a complex terrain with sunlight reflection, there may be various light wavelength interferences. Therefore, clarifying the main frequency component helps to optimize the basis for subsequent adjustment algorithms. The main frequency component here is usually obtained by Fourier transform or other decomposition methods, and can represent the maximum power distribution in an ideal situation.

[0069] The second step involves calculating the spectral distribution variance formula: σ2 = 1 / N ∑(f_i - f_mean)2, where N represents the number of different frequency points obtained by sampling, and the range can be very large depending on the resolution and detection distance, while the optimal value depends on the maximum bandwidth supported by the specific hardware; f_i refers to the absolute energy value or signal intensity value at the i-th frequency, and these measurement data are input into the system after preprocessing such as filtering and noise removal; f_mean is the average of all f_i to reflect the overall homogenization level. The design of this formula aims to understand the changes in ambient light conditions through variance measurement. If the variance is too large, it indicates that there are obvious inconsistencies or abnormal sources that need to be identified. The variance is selected as the key statistical indicator because it is sensitive to fluctuations and can well express heterogeneous information.

[0070] Next, when the calculated variance σ2 exceeds a given threshold threshold1 (this value needs to be set according to experimental results), the third step is performed, which is to reset the sensor sensitivity. The new value of the sensitivity is defined in the form of k×σ2, where k is the sensitivity adjustment factor, and the default value is generally set within a certain range to ensure reasonable gain and avoid saturation. Such a setting enables the automatic amplification of weak signals and the suppression of overly strong changes in a high-difference environment, ensuring that the stable operating performance of the entire system is in a good state.

[0071] The fourth and final step focuses on the sensor calibration process, which is a regular correction action implemented to maintain consistent response characteristics and accuracy among different devices during long-term use. In one embodiment, if a tracked robot is passing through a mixed area of urban streets and woods with complex vegetation-reflected light, it first captures a full-spectrum image containing multiple characteristics such as the sky background and ground material reflections, and extracts the main frequency information such as the strong absorption characteristics corresponding to the green band. At the same time, the system detects a relatively high σ2 indicating significant spatial diversity, so it increases the parameter weights of the camera module and other sensing units for distinguishing approximately specific color objects according to the rules, and then verifies whether the output conforms to the pre-stored standard model curve and makes fine-tuning. In this way, reliable monitoring support can be continuously provided regardless of external challenges.

[0072] The present invention also includes performing panoramic vision fusion processing on the collected data, and this process includes four main steps: calculating the scene illumination deviation, global brightness compensation, panoramic stitching, and edge-preserving filtering.

[0073] The first step is to calculate the scene illumination deviation according to the image pixel distribution by the formula ΔL = |I_max - I_min| / I_ave, where I_max and I_min represent the maximum brightness value and the minimum brightness value in the image respectively, and I_ave represents the average brightness value. ΔL reflects the degree of brightness change in the image. Usually, the value range of I_max is [0, 255] (taking an 8-bit image as an example), and the value of I_ave depends on the average brightness of the specific image. For example, in a strong light environment or under shadow occlusion, I_max is significantly higher than I_min, which will cause the value of ΔL to increase, thus reflecting the imbalance of the scene illumination distribution.

[0074] If ΔL exceeds the pre-set threshold threshold2, the second step will be initiated: applying the global brightness compensation formula C_compensate = C_original + λ × ΔL to enhance the image, where C_original is the color value of the original image, and λ is the compensation factor used to adjust the degree of image enhancement, usually selected in the range of [0.5, 1.5]. When ΔL is too high, using this formula can balance the brightness of the image, so that the shadow part is enhanced while preventing overexposure. The optimal value of λ is usually set to 1 because it can achieve an ideal enhancement effect under different illumination conditions.

[0075] The third step uses a panoramic stitching algorithm to seamlessly integrate the images into a 360° view. Specifically, by matching and optimizing the image feature points, the boundary influence of the overlapping area is eliminated to form a complete three-dimensional panoramic view. This process is applicable to data fusion in complex environments and ensures the consistency of the generated view.

[0076] The fourth step eliminates artifacts through edge-preserving filtering technology to enhance the detail quality in complex environments. This method can effectively suppress noise interference and protect the texture and structure information of the original image.

[0077] In one embodiment, assume that a tracked robot is performing an intelligent reconnaissance mission outdoors, and the scene includes complex multi-light source conditions such as direct sunlight areas, shadow areas, and grassland backgrounds. Initially, the robot collects a series of image data with large light and dark contrasts through a multispectral camera, and the calculated ΔL value shows that it significantly exceeds threshold2. Therefore, the system automatically executes the global brightness compensation step to adjust the visibility of the shadow area and optimize the display effect of the strong light area. Subsequently, the panoramic stitching connects the views collected in different directions to form a continuous and complete display of the surrounding environment. Finally, the edge-preserving filter further improves the authenticity and usability of the image, making the reconnaissance result more reliable and clear, meeting the requirements of the actual application scenario.

[0078] The present invention also includes performing multispectral data analysis based on the fusion data. The analysis process mainly includes the following steps: normalizing the energy values of different bands, defining the spectral matching degree of the target substance, selecting the reference material with the optimal matching degree as the material type, and adding a probability classification module to adapt to uncertainties.

[0079] The first step is to process the energy values of different bands using a normalization algorithm. The formula is E_norm(i) = (E_i - E_min) / (E_max - E_min), where E_i represents the original energy value of the i-th band, and E_min and E_max respectively represent the minimum and maximum energy values among all bands. The value range is usually the set of non-negative real numbers. This formula compresses the energy values into the range of 0 to 1 through a linear stretching method. The purpose is to remove the influence of dimensions, make the data comparable, and improve the calculation efficiency. For example, when a tracked robot is performing reconnaissance and the energies of the infrared and ultraviolet bands collected are 52 units and 300 units respectively, normalization unifies the two to the same scale for comparison.

[0080] The second step involves defining the spectral matching degree D of the target substance, which is completed through the formula D = Σ|S_ref(i) - E_norm(i)|, where S_ref(i) is the spectral data of the reference material and i is the index of the corresponding band. The logic of this formula is to accumulate and sum based on the difference between the reference value and the actual detection value of each band. The greater the difference, the higher the matching degree, and the less ideal the D value. Through this operation, the most similar reference sample can be found. Suppose the reference spectral values of each band of a certain material sample are [0.7, 0.3], and the measured normalization results E_norm are [0.68, 0.28] respectively. Then the smaller error between them indicates that the substance may be close to the known sample type.

[0081] Based on the above, the third step is to select the category with the smallest matching degree D as the recognition target. The core of this step is to compare multiple possible S_ref objects and select the best approximation. If there are three types of reference materials marked as type A, type B, and type C respectively in a group of candidate reference material types, and their calculated D values are 0.2, 0.15, and 0.13 respectively, then type C is finally confirmed as the material to which the target belongs. This is because in the assumption, C has the lowest overall deviation sum.

[0082] The fourth step introduces a probabilistic classification mechanism to handle possible data errors, environmental interferences, and other situations. Even if the attribution of a primary material is determined, it is necessary to quantify the possibility of other potential candidate materials. In one embodiment, when the tracked robot conducts intelligent reconnaissance work under uneven light intensity or heavy occlusion, even if it is initially determined that the metal material composition mainly comes from the aluminum alloy series, a probability label needs to be added to indicate the extremely low proportion of ferromagnetic elements or the probability of the existence of other unknown dopants. Specifically, this additional step ensures that the analysis is more robust, thus meeting the complex requirements in real application scenarios.

[0083] The present invention also includes a correlation optimization algorithm that combines visual data and spectral information, including the following steps: First, construct a joint feature space; second, calculate the weight parameters through covariance analysis; third, set an interference rejection threshold for filtering invalid data; and finally, retrain the classifier based on the remaining high-confidence features to improve the target recognition accuracy.

[0084] The first step is to construct a joint feature space J = [V1, V2,..., VS], where Vi represents the position of each target in this feature vector space. This operation aims to integrate the information of panoramic vision and multi-spectral dimensions into a unified mathematical structure for subsequent comprehensive analysis. Each dimension Si is defined as the mapped coordinate value of the target in its respective observation dimension, and the range is determined according to the specific sensor characteristics of the acquisition. For example, in some application scenarios, Vi can be a normalized numerical vector from 0 to 1 to ensure that all inputs are consistent for unified processing. The constructed feature matrix enables the efficient utilization of information from multiple data sources and also provides a clear and quantifiable way to represent the target.

[0085] The second step is to use covariance analysis to determine the importance degree of each feature component to the final classification result. Its formula expression is w_j = Σcov(F_vis(j), F_spec(j)) / Σvar(F_vis(j)), where F_vis and F_spec respectively refer to the visual and spectral feature vector sequences extracted from the same scene but separately; cov is used to measure the co-variation trend, that is, the linear correlation degree, between the characteristics of two different sources, and var represents the internal fluctuation level or discrete intensity of a single source. The weight parameter w_j represents the evaluation index of the contribution degree of the visual-light fusion data to the overall system decision support under dimension j. Usually, its optimal range should be between -1 and +1. The closer it is to a positive number, the more significant the consistency and effectiveness of the information provided by the corresponding channel; if the absolute value is low or negative, it indicates that the two may have low collaborative potential or there is a risk of misguidance.

[0086] On the above basis, a criterion for removing interference items is set, and a fixed constant threshold4 is introduced to compare with the actual weight value. When the inequality condition w_j < threshold4 is met, it indicates that the element may be a non-critical or false signal and needs to be excluded. Here, it is recommended that the threshold4 value is generally selected as a boundary close to zero small positive, such as 0.2, to ensure sufficient sensitivity without being overly conservative. In one embodiment, it is assumed that a large set of samples {w} mixed with noise is obtained in the initial stage. After calculation and verification, it is found that the parts below the threshold are generally caused by errors such as equipment hardware aging and external strong light pollution. By strictly restricting these interferences, a higher-purity data group can be obtained for the next step.

[0087] The final step is to use the high-quality attributes after screening to perform the iterative training process of the classification model again to enhance the accuracy output level of the detection efficiency. Specifically, in a panoramic reconnaissance case, assume that the task of a tracked robot is to distinguish normal vegetation from camouflaged objects in the ground environment. The key indicators (such as the matching frequency of color-uniform regions and the performance of short-wave infrared absorption incident differences) retained after the screening in the first three stages will become the key input materials for the new algorithm to recalibrate the prediction logic framework of the SVM support vector machine, and finally, in the actual combat test in complex terrain, it shows a significant improvement in the occurrence probability of misreporting that was prone to occur before, and significantly enhances the adaptability and practicality of the system.

[0088] The filtering of interference in multi-target reconnaissance of the present invention is further limited to: this process includes multiple steps to achieve the goal of optimizing reconnaissance efficiency. In the first step, the interference intensity signal collected in real time is mapped to the interference level index \(M), and the specific formula is \(M = \log(\sum|I_{\text{noise}} - I_{\text{baseline}}|^2) / \log(N_{\text{noise}}))). In this formula, \(I_{\text{noise}}) represents the interference data value monitored in real time, \(I_{\text{baseline}}) is the reference value of the baseline interference, which represents the expected environmental interference standard of the system, and generally ranges in the normalized interval between -1 and 1, and its optimal setting depends on the adjustment of experimental data; \(N_{\text{noise}}) represents the total number of interference frames, that is, the number of frames of the pictures detected as being interfered within a unit time, and the range is a positive integer. The design logic of the formula is to measure the energy size of the interference by summing the squares of the signal differences, and to smooth the output of the interference level index \(M) by taking the logarithmic ratio relationship with the total number of interference frames. The main function of this step is to extract a comparable interference magnitude evaluation value from the original data.

[0089] In the second step, after obtaining the interference level index \(M\), check whether the threshold condition \(M > threshold5\) is satisfied. If so, start the redundant data verification logic to remove invalid frames. Invalid frames refer to fragments of image information that deviate significantly from the benchmark. For example, when the robot is in a sandy environment, strong sand reflections may cause some images to be unrecognizable, and they need to be marked for deletion or ignored in the calculation.

[0090] In the third step, for the data within the extracted valid frames, start analyzing the relative motion characteristics \(H_{text{diff}}(j)\) between multiple targets, where each \(j\) represents a set of target position change parameters at different time points. This step is used to distinguish stationary and moving entities. Suppose a certain target has a large displacement relative to the background pixels, it may be determined as a moving target rather than a noise source, and thus further in-depth analysis is carried out in the next step.

[0091] The last step is to prioritize focusing on high-value moving targets based on the previous classification results, thereby further improving the reconnaissance efficiency. In one embodiment, if it is detected by panoramic vision fusion multi-spectral technology that an enemy tank is moving slowly (judged as a moving type), while static objects by the roadside or the swaying of grass leaves caused by the wind are not regarded as potential threats, resources will be concentrated on processing the dynamic trajectory capture of key military equipment. Specifically, the tank has a significant strategic significance score, so more processor computing power is allocated for trajectory prediction and warning generation.

[0092] The optimized interference conditions of the present invention include four steps, namely updating the evaluation formula of the interference level, starting the standby sensing system, integrating redundant sensing data to form a closed-loop correction structure, and improving the target positioning resolution in the key frame detection stage.

[0093] The first step is to introduce a self-learning adjustment module to update the evaluation formula of the interference level to \(A = \alpha M+\beta\sum H_{text{rel}}\), where the parameters \(\alpha\) and \(\beta\) represent weights respectively, and their values range within [0,1]. \(M\) represents the basic interference level measured in the current multi-spectral analysis, and \(H_{text{rel}}\) is the sum of relative relationship characteristics, which usually reflects the strength of the correlation between multiple sensing data. The optimal values of \(\alpha\) and \(\beta\) need to be verified through experiments. For example, when the environmental complexity increases, \(\alpha = 0.6,\beta = 0.4\) can be set. The meaning of this formula is to calculate the interference level \(A\) by combining the basic interference amount and relative relationship characteristics to achieve a more accurate assessment of the actual interference degree. The main purpose of setting the formula is to enhance the comprehensive understanding of the interference conditions by combining basic information and the relationship between data.

[0094] The second step is that if the calculated interference level \(A > \text{threshold}_6\), the backup sensing system is enabled. Here, \(\text{threshold}_6\) is a pre-set critical value, which aims to enable redundant sensing hardware in a timely manner when the interference conditions exceed a specific limit to ensure that the robot can still maintain stable detection performance in a complex environment. For example, in one embodiment, the threshold \(\text{threshold}_6\) can be set to the empirical value 35 (the unit varies according to the definition). When the result of the evaluation formula exceeds this value, the robot switches to operate with a higher-priority sensor group to avoid functional failures caused by data errors.

[0095] The third step involves combining redundant sensing data with the existing reconnaissance model to construct a closed-loop correction mechanism. This step aims to effectively utilize different redundant signals collected by the sensors, thereby adjusting the model output in real time and improving the accuracy and reliability of overall intelligent reconnaissance. Specifically, in the scenario of panoramic vision and multispectral analysis, by integrating thermal imaging and ordinary visible light images to generate a more robust data set, the model's ability to identify environmental features can be dynamically optimized.

[0096] Finally, the target localization resolution in the key frame detection stage is improved to \(R_{\text{loc}} \geq \text{resolution}_{\text{goal}}\). This goal is to enhance the robot's precise positioning of specific objects when dealing with important reconnaissance tasks and reduce the adverse effects of errors on task completion. In one example, it is required that the resolution \(R_{\text{loc}}\) reaches at least 0.2° (the angular resolution reaches the specified accuracy target). Such a design ensures that even in the face of distant or fast-moving targets, reliable tracking means are available to meet the requirements of high-precision map construction and tactical deployment.

[0097] For the real-time feedback correction path planning of the present invention, the definitions and implementation processes of each step are first clarified: defining the deviation ratio, setting the trigger conditions, generating candidate paths, screening the optimal paths, and updating the prediction ability. The following will elaborate on these steps in turn.

[0098] First, define the proportional deviation formula for the current position deviating from the reference trajectory as \(P_{text{deviation}} = |pos_{text{current}} - pos_{text{benchmark}}| / path_{text{len\_total}}\). The parameter explanations for this formula are as follows: \(pos_{text{current}}\) represents the current actual position of the robot, \(pos_{text{benchmark}}\) is the desired position on the preset reference path, and \(path_{text{len\_total}}\) is the total length of the entire path. The role of the formula is to quantify the degree of the robot's trajectory deviation, and its value range generally lies between [0, 1]. The optimal value is dynamically adjusted according to the task requirements. This indicator is used to evaluate whether the deviation from the reference trajectory is excessive.

[0099] Second, if the deviation ratio \(P_{text{deviation}} > threshold7\), or the detected distance to the obstacle \(D_{text{obst}} \leq threshold8\), then start the replanning program. Here, \(threshold7\) is the maximum allowable deviation threshold set manually, usually set around 0.2; \(threshold8\) is the minimum obstacle safety distance to avoid risks, generally set as twice the radius of the robot arm's extension or the width outside the crawler, such as about 1 meter. Setting such trigger conditions can ensure that path deviations or potential collisions are corrected in a timely manner.

[0100] Third, generate multiple candidate routes \(r_{text{candidate}}\). In this process, it is necessary to combine environmental information (panoramic vision, multispectral analysis data) and kinematic constraints to create an executable path set. Under each candidate route, there are different cost estimates such as time, space, and energy consumption.

[0101] Fourth, screen the paths according to the estimated cost function \(COST(r_{text{candidate}}, t_{text{now}})\). Among them, the \(COST\) function comprehensively considers factors such as the path length, estimated arrival time, difficulty of avoiding obstacles, and energy consumption, preferably selects the candidate path with the lowest cost to perform operations, and retains relevant historical data for subsequent iterative improvement of the model performance.

[0102] In one embodiment, assume that a tracked robot is moving along a predefined route during a mountain reconnaissance mission. Due to the irregular terrain, the panoramic camera detection shows that the robot has deviated to the right from the original trajectory by approximately 0.25 times the total path length ((P_{text{deviation}})), exceeding the set threshold ((threshold7 = 0.2)). Additionally, the front sensor indicates that the distance to an obstacle is less than 0.9 meters ((threshold8 = 1)). When either of the above two conditions is met, the path re-planning process is automatically entered. At this time, the potential surrounding paths are extracted through multi-spectral image processing and the cost is estimated. Finally, choosing to detour to the left as the new instruction and outputting it to the control unit to continue the mission advancement. In this way, the goal of the intelligent feedback correction function is specifically achieved.

[0103] The further enhanced path intelligent correction step of the present invention is as follows: First, a prediction window is established, and the conflict point set is screened through a judgment formula for avoidance. Secondly, a two-level obstacle avoidance rule is applied for adjustment. Finally, after the adjustment is completed, it is detected again to ensure that the standard condition exit_flag = True is met.

[0104] The steps are as follows: The first step is to define the time prediction window T_future as the product of the remaining time and a magnification factor γ (T_future = time_remain * γ), which is used to evaluate the potential problem points P_future that may be encountered before the end of the robot mission. The parameter time_remain represents the remaining time required to complete the current reconnaissance mission; the parameter γ represents the time proportionality coefficient based on the task urgency, and its value range is usually set between 0.8 and 2. The preferred value of γ is taken as 1.5. In the case of high priority, a higher γ value is set to extend the future prediction time to cover more scenario complexities. The function of this formula is to generate a reasonable future analysis period to make the robot's planning more predictable. In one embodiment, if the task requires rapid acquisition of panoramic intelligence data, then γ is set to a larger value such as 1.8, which can extend the robot's advance prediction of possible situations.

[0105] The second step is to apply the conflict judgment formula C(p_i, t_estimated, v_target) > critical_conflict_value to all points within the problem point P_future to identify the points that need to be specifically avoided. In this formula, p_i is each individual problem position point in the set, t_estimated is the estimated time for the target position p_i to be reached, v_target is the target forward speed, and critical_conflict_value is a predefined threshold. The purpose of this step is to confirm whether there are obstacles or task constraints that make the route unsafe. For example, when a tracked robot is crossing a terrain covered with weeds and gravel, once it is calculated that certain specific coordinates p have a risk score exceeding the critical conflict value, it is considered that there are unacceptable hazards in these places and they need to be avoided.

[0106] Then a two-level obstacle avoidance strategy is adopted, that is, locally fine-tuning the azimuth angle of the robot's head to avoid obstacles, and then re-regulating the traveling direction from the overall direction to prevent excessive detours and increase unnecessary distance consumption. For example, when facing a small mound blocking on the right front, the vehicle's direction can be slightly changed first, swinging slightly to the left to get out of the edge of the obstacle area and then resuming the normal path to continue moving forward. This kind of processing can not only effectively get out of danger but also minimize the total driving distance to achieve high efficiency.

[0107] The last link is to review the entire process output again after all the above corrective actions are completed to check if the exit_flag flag is correct and all meet the ideal state. Specifically, during a patrol mission, when it is found that a certain path segment does not meet the final safety evaluation criteria, the above-mentioned several corrective and optimization links are repeatedly looped until the entire path is in an ideal condition. This repeated verification step ensures the practical usability and robustness of the final solution.

[0108] The present invention improves the multi-camera synchronization efficiency, and this process mainly includes the following steps: setting the trigger delay of the camera synchronization signal, monitoring the frame rate change and giving early warnings and adjustments, ensuring that the imaging quality uniformity error meets the tolerance requirements, and achieving high-efficiency response and energy consumption optimization.

[0109] First, set the camera synchronization signal trigger delay to control the stability of the actual delay range through the formula \(\tau_{sync\_actual}=\lceil\Delta\tau / step\_unit\rceil\). In the formula, \(\Delta\tau\) is the camera asynchronization time error in the current system; \(step\_unit\) is the unit step parameter, which is used to quantify and classify the error. Usually, its value ranges from 1 to 5 ms, and the optimal value is set according to the application scenario. For example, a small step is preferred during rapid reconnaissance to reduce jitter errors. This formula defines the actual delay \(\tau_{sync\_actual}\) of the synchronous trigger as the ceiling result, aiming to make the trigger timing fall within a stable time window. For example, in a multi-camera system, if the maximum asynchronization error \(\Delta\tau\) is detected to be 7 ms and the unit step is set to 2 ms, the calculated trigger delay by the formula is 4 ms (quantified to the nearest integer multiple), which can effectively stabilize the synchronization relationship between multiple cameras.

[0110] Second, when the change in the frame rate exceeds the allowable range \(\delta fps=|fps_{current}-fps_{designed}| / fps_{designed}\) and exceeds the preset threshold threshold_fps, a warning is triggered and the light source intensity is adjusted. In this formula, fps_current and fps_designed are the real-time measured and pre-designed frame rates respectively, and the difference between the two is normalized to a relative error to define whether measures need to be taken. When the error exceeds the threshold, it indicates that there may be light instability or other environmental disturbances during the multi-spectral imaging process. At this time, the light source intensity is adjusted to restore the frame rate stability, thereby ensuring the effectiveness of the subsequent processed data. For example, in a complex outdoor environment, due to unstable light sources, the current frame rate is 25 fps (the designed target frame rate is 30 fps), that is, \(\delta fps=0.167>threshold\_fps\) (set to 0.1), so the aperture will be adjusted or an external supplementary light source will be added to improve the problem.

[0111] In addition, it is necessary to ensure that the imaging quality balance error err_quality of all channels does not exceed the set tolerance interval_tolerance. Here, err_quality includes evaluations in aspects such as color saturation, resolution, and contrast, and standardized comparisons of different spectral imaging are required. If the imaging of some channels is unbalanced, it may affect the effect of the panoramic image after multi-spectral fusion. Specifically, in an embodiment, assume that the clarity of the red visible light sensor drops by 10% under low light conditions, while the clarity of the infrared sensor only drops slightly by 3%. Then, the former should be optimized separately to reduce the quality difference to within the tolerance range, such as using image post-processing means to improve the signal-to-noise ratio.

[0112] Finally, the above strategies should minimize energy waste while ensuring high system performance. For example, dynamic voltage and frequency scaling technology can be used to reduce power consumption, or intelligent power scheduling can be adopted to allocate loads to the main functional modules. These measures help the tracked robot maintain its mission operation ability for a longer time during the reconnaissance process and are applicable to long-term operation scenarios in complex terrains and adverse weather conditions.

[0113] The enhanced image quality problem repair step of the present invention consists of four parts: recording parameters, applying a histogram correction formula, integrating a deep learning model, and loop verification.

[0114] First, record the global exposure parameter `; exp_global`; and the local brightness ratio `; factor_bright_localavg_intensity_high / avg_intensity_low`;. This step captures the information of high-contrast regions by quantifying the global and local brightness features. Among them, `; exp_global`; represents the overall exposure compensation coefficient in the current scene (usually in the range of 0.1 to 2), and `; factor_bright_local`; is calculated based on the ratio of the average intensity of the local bright area `; avg_intensity_high`; to the average intensity of the dark area `; avg_intensity_low`; and is used to reflect the local illumination distribution difference. This process can provide a data basis for subsequent processing, such as determining the compensation benchmark when the lighting conditions change suddenly.

[0115] Second, use the histogram correction formula `; hist_new(z) = exp_global × factor_correction × factor_bright_local`; to optimize the brightness distribution for high-contrast regions. Where `; z`; is the position variable of the image pixel intensity value, and `; factor_correction`; is a correction constant (it is recommended to set the range from 0.5 to 1.5, and the optimal value is 1 to adapt to general scenarios). By adjusting the histogram, the intention of this formula is to map the extreme brightness intervals to a more moderate range. Since high contrast may be caused by complex natural lighting or artificial interference, in one embodiment, for the problem of overexposure of the image when the tracked robot enters a semi-light and semi-dark tunnel environment, the formula ensures that the details of the image are not lost.

[0116] Third, integrate the trained deep learning model to optimize the exposure and sharpness of the initial image `Q_initial`, and generate the final optimized image `Q_optimize = net(Q_initial, fine_tuning_params)`. At this stage, the fine-tuning parameter `fine_tuning_params` defines the behavior of the model during operation, and its value can be customized according to specific tasks, such as increasing the weight ratio of the infrared spectral band. Specifically, combined with the multi-spectral fusion characteristics, the model can significantly improve the edge sharpness and compensate for information distortion. For example, during the reconnaissance process of a tracked robot, if the image quality deteriorates due to smoke and dust, this step can achieve detail restoration through intelligent analysis.

[0117] Fourth, perform loop closure verification regularly to check the performance consistency of the above method during long-term use. The core objective of this step is to ensure that the algorithm performance does not degrade with changes in scene conditions or device status. By iteratively evaluating the output results under different lighting environments, such as comparing whether the recognition accuracy of the robot between day and night alternation scenarios meets the standards, ensure the stability and reliability of the method.

[0118] The intelligent reconnaissance system of the tracked robot based on panoramic vision fusion and multi-spectral analysis of the present invention includes:

[0119] First, to solve the problem of decreased reconnaissance accuracy under specific conditions, this method dynamically adjusts various sensor parameters by analyzing the environmental spectral characteristics (such as different lighting, weather, or occlusion conditions), such as the exposure time, gain value, and acquisition frequency of panoramic vision and multi-spectral data. This dynamic adjustment mechanism ensures that the sensors can always maintain sensitivity and accuracy in complex environments, thus providing high-quality data input for subsequent data analysis.

[0120] Second, for the image quality problem caused by light changes, this method adopts advanced panoramic vision fusion processing technology. By performing fusion operations such as registration, denoising, and enhancement on data from multiple perspectives and spectral bands, effectively reduce the interference caused by factors such as uneven light and shadows. This method can compensate for the negative impact brought by light changes in complex environments and improve image quality and information integrity.

[0121] Third, during the target material classification process, extract the unique characteristics of the object's reflection spectrum through multi-spectral data analysis, and perform precise comparison in combination with the pre-established material library model to complete the target material classification. This method based on multi-spectral data analysis overcomes the problem of confusing material recognition in traditional single-image reconnaissance means and greatly improves the accuracy and robustness of reconnaissance.

[0122] Furthermore, to address the interference problem in multi-objective reconnaissance, the present invention constructs an optimization algorithm framework based on the strong correlation between visual data and spectral information to extract and enhance target feature information while filtering background interference signals, enabling the system to focus on key mission objectives in real time. By reducing the weight or filtering out non-relevant information, the performance stability in multi-objective scenarios is significantly improved.

[0123] Finally, the method also has the characteristic of path planning based on real-time reconnaissance data feedback. Specifically, by combining machine learning and prediction algorithms, it analyzes the current target reconnaissance data and historical path records to dynamically correct the robot path. This enables the tracked robot to still timely track and avoid obstacles or misjudgment points even when the target position changes, reducing the risk of target loss, thereby ensuring the success rate and efficiency of the reconnaissance mission. In summary, this reconnaissance method comprehensively improves the adaptability and reconnaissance ability of the tracked robot from multiple dimensions.

[0124] Although the embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the invention, and the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A tracked robot intelligent reconnaissance system based on panoramic vision fusion multi-spectral analysis, characterized in that: include: Dynamically adjust sensor parameters based on environmental spectral characteristics to adapt to different light conditions and collect high-quality panoramic vision and multispectral data; Perform panoramic visual fusion processing on the collected data to compensate for light changes in complex environments and enhance image quality; Perform multispectral data analysis based on fused data, and classify target materials based on the analysis results to improve reconnaissance accuracy; combine the correlation optimization algorithm of visual data and spectral information to filter out interference in multi-target reconnaissance and provide real-time feedback to correct the robot's path planning; The dynamically adjusting the sensor parameters based on the environmental spectrum characteristics further comprises: determining the main frequency component f_main of the current environmental spectrum; Calculate the spectral distribution variance σ2=1 / N∑(f_i-f_mean)2, where N is the number of frequency points, f_i represents the intensity of the i-th frequency point, and f_mean is the average frequency intensity; when σ2>threshold1, adjust the sensor sensitivity to k×σ2, where k is the sensitivity coefficient; calibrate the sensor to ensure the consistency and accuracy of the data.

2. According to claim 1, a tracked robot intelligent reconnaissance system based on panoramic vision fusion multi-spectral analysis is characterized in that: The panoramic visual fusion processing of the collected data includes: According to the image pixel distribution, the scene illumination deviation ΔL=|I_max-I_min| / I_ave is calculated, where I_max and I_min are the maximum and minimum brightness values, respectively, and I_ave is the average brightness value; If ΔL exceeds the threshold threshold2, the global brightness compensation formula C_compensate = C_original + λ × ΔL is used for image enhancement, where λ is the compensation factor; Apply panoramic stitching algorithms to seamlessly stitch images into a 360° view; Use edge-preserving filtering to remove artifacts and enhance detail quality in complex environments.

3. According to claim 1, a tracked robot intelligent reconnaissance system based on panoramic vision fusion multi-spectral analysis is characterized in that: The multispectral data analysis according to the fused data comprises: The energy values ​​E_i of different bands are processed by normalization algorithm to obtain E_norm(i)=(E_i-E_min) / (E_max-E_min); Define the target material spectrum matching degree D = Σ|S_ref(i)E_norm(i)|, where S_ref(i) is the reference material spectrum data; Select the category of S_ref(i) corresponding to the smallest D as the material type; Added a probabilistic classification module for material types to accommodate uncertainty.

4. According to claim 1, a tracked robot intelligent reconnaissance system based on panoramic vision fusion multi-spectral analysis is characterized in that: The correlation optimization algorithm combining visual data and spectral information includes: Construct a joint feature space J = [V1, V2, ..., VS], where Vi represents the position of each target in the feature vector space; The weight w_j = Σcov(F_vis(j), F_spec(j)) / Σvar(F_vis(j)) is obtained through covariance analysis, where F_vis and F_spec represent the variance of the visual and spectral feature dimensions respectively; Set the interference removal threshold threshold4 and remove the interference w_j according to the following conditions <threshold4; The classifier is retrained based on the remaining high-confidence features to accurately identify the target.

5. According to claim 1, a tracked robot intelligent reconnaissance system based on panoramic vision fusion multi-spectral analysis is characterized in that: The filtering of interference in multi-target reconnaissance also includes: Map the interference intensity signal collected in real time to the interference level index M, M = log(∑|I_noise I_baseline|^2) / log(N_noise), N_noise represents the total number of interference frames; When M>threshold5, the redundant data verification logic is started to remove invalid frames; Extract the relative motion feature H_diff(j) between multiple targets in the valid frame to distinguish between stationary and moving entities; Based on the classification results, high-value moving targets are prioritized to optimize reconnaissance efficiency.

6. The tracked robot intelligent reconnaissance system based on panoramic vision fusion multi-spectral analysis according to claim 5 is characterized in that: The optimized interference condition is: The evaluation formula for updating the interference level by introducing the self-learning adjustment module is A=αM+β∑H_rel, where α and β are weight parameters and H_rel is the sum of relative relationship features; If A>threshold6, the backup sensing system is enabled to ensure reliability; Integrate redundant sensor data into the existing reconnaissance model to form a closed-loop correction structure; Improve the target positioning resolution R_loc ≥ resolution_goal in the key frame detection stage.

7. The tracked robot intelligent reconnaissance system based on panoramic vision fusion multi-spectral analysis according to claim 1 is characterized in that: The real-time feedback correction path planning includes: Define the proportional deviation of the current position from the benchmark trajectory as P_deviation = |pos_current pos_benchmark| / path_len_total; When P_deviation>threshold7 or obstacle distance D_obst≤threshold8, the re-planning procedure is started; Generate multiple candidate routes r_candidate and filter them according to the estimated cost function COST(r_candidate,t_now); The path with the lowest cost is selected preferentially while updating historical data to support improved predictive capabilities.

8. The tracked robot intelligent reconnaissance system based on panoramic vision fusion multi-spectral analysis according to claim 7 is characterized in that: The steps of enhancing the intelligent correction of the path are as follows: Establish a prediction window T_future = time_remain*γ (the time window multiplier γ is set according to the task priority) to predict the problem point set P_future that may be encountered in the future; If any p∈P_future satisfies the conflict judgment formula C(p_i,t_estimtated,v_target)>critical_conflict_value, avoid it in advance; Introducing a dual-level obstacle avoidance rule to first adjust the angle locally and then adjust the direction overall to avoid excessive detours; After the adjustment is completed, the safety check is performed again until the standard condition exit_flag = true is met.

9. The tracked robot intelligent reconnaissance system based on panoramic vision fusion multi-spectral analysis according to claim 1 is characterized in that: Improving the efficiency of multi-target synchronization includes: Set the camera synchronization signal trigger delay τ_synchronization and use the formula τ_sync_actual = ceil[Δτ / step_unit)] (step_unit is the unit step distance) to control the actual delay range to be stable; When the frame rate changes beyond the allowable range δfps = |fps_current fps_designed| / fps_designed is greater than threshold_fps, the warning mechanism is triggered and the light source intensity is adjusted; Ensure that the image quality balance error err_quality of all channels does not exceed the set tolerance interval_tolerance; Achieve fast response while reducing energy consumption to adapt to long-term operating environment requirements.

10. The tracked robot intelligent reconnaissance system based on panoramic vision fusion multi-spectral analysis according to claim 1 is characterized in that: The steps to fix the enhanced image quality issues include: Record the global exposure parameter exp_global and the local brightness ratio factor_bright_local = avg_intensity_high / avg_intensity_low; For high contrast areas, the histogram correction formula hist_new(z)=exp_global×factor_correction×factor_bright_local is used; Integrate deep learning model to fine-tune exposure and sharpening effect to output optimized image Q_optimize = net (Q_initial, fine_tuning_params); Regularly loop back to verify the effectiveness of the repair strategy to ensure long-term consistent and stable performance.

Images