A mycelium topology-oriented image super-resolution reconstruction method

Abstract

The present application relates to the technical field of mycelium image processing, and discloses a mycelium topological structure-oriented image super-resolution reconstruction method. The method acquires a low-resolution topological image of a mycelium sample and extracts an initial resolution matrix; based on a preset mycelium structure feature library, the initial matrix is subjected to topological structure correction to generate a first corrected resolution matrix; the first corrected resolution matrix is subjected to dynamic enhancement by a mycelium growth simulator to output a target resolution matrix. Meanwhile, an actual topological image is collected to extract an actual resolution matrix; when the structural difference degree of the two exceeds a threshold value, a first deviation matrix is generated and the target matrix is iteratively adjusted to obtain a first adjusted matrix. The adjusted matrix is taken as a new target matrix to repeat the dynamic enhancement step until the structural difference degree is less than the threshold value, thereby realizing high-precision mycelium topological image reconstruction. The method improves the accuracy of mycelium topological image super-resolution reconstruction through feature library correction, dynamic enhancement and iterative adjustment.

Description

Technical Field

[0001] This invention relates to the field of mycelial image processing technology, specifically to an image super-resolution reconstruction method oriented towards mycelial topology. Background Technology

[0002] The structural characteristics analysis of mycelium is of great value in many research fields, including microbiology, agricultural science, and materials science. As the basic unit of fungal vegetative cells, the topological structure of mycelium (such as branching patterns, connection methods, and growth direction) directly reflects the growth state, metabolic activity, and interaction with the environment. However, due to the microscopic, delicate, complex structure and susceptibility to external environmental interference, obtaining high-quality mycelial topological images using conventional microscopic imaging equipment presents many challenges.

[0003] Image super-resolution reconstruction techniques are widely used to improve the detail of low-resolution images, but specialized super-resolution methods for mycelial topology have significant shortcomings. Traditional super-resolution methods are mostly based on pixel-level enhancement of general images, often focusing on improving the overall image sharpness while neglecting the preservation of topological features of specific biological structures like mycelia. When processing low-resolution mycelial images, these methods are prone to topological distortion, such as misjudging branch points, blurring connectivity, or distorting growth directions, making the reconstructed image unable to accurately reflect the true topological morphology of the mycelia.

[0004] Noise interference and loss of detail are common problems in the initial resolution matrix extraction process of low-resolution mycelial images. Existing correction methods mostly rely on general image denoising or smoothing algorithms, lacking specific consideration for the structural features of mycelia. This can lead to further loss of key topological information during the correction process, making it difficult to generate a reliable initial processing foundation. Furthermore, mycelial growth is a dynamic process, and its topological structure changes regularly over time. Traditional static super-resolution methods cannot simulate this dynamic growth characteristic, and the reconstruction results are difficult to reproduce the true topological evolution of mycelia under natural growth conditions.

[0005] Current technologies lack effective feedback and adjustment mechanisms. After reconstruction, it is often impossible to identify and optimize deviations through precise comparison with actual high-resolution images, leading to potential significant differences between the reconstructed results and the actual hyphal topology. These differences directly impact subsequent structural analysis and parameter extraction, reducing the reliability of conclusions drawn from the reconstructed images. Therefore, developing a super-resolution reconstruction method that accurately preserves topological features, dynamically simulates the growth process, and possesses self-optimization capabilities, specifically tailored to the unique characteristics of hyphal topology, has become a pressing issue in this research field. Summary of the Invention

[0006] The purpose of this invention is to provide an image super-resolution reconstruction method oriented towards mycelial topology, so as to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, this invention provides an image super-resolution reconstruction method oriented towards hyphal topology, the method comprising:

[0008] Obtain a low-resolution topological image of the mycelial sample, and extract the initial resolution matrix of the low-resolution topological image;

[0009] The initial resolution matrix is ​​topologically corrected based on a preset mycelial structure feature library to generate a first corrected resolution matrix;

[0010] The first corrected resolution matrix is ​​dynamically enhanced using a mycelial growth simulator, and the target resolution matrix is ​​output.

[0011] The actual topological image of the mycelial sample was acquired using a high-precision microscopic imaging device, and the actual resolution matrix was extracted from the actual topological image.

[0012] When the structural difference between the target resolution matrix and the actual resolution matrix exceeds a preset difference threshold, a first deviation matrix is ​​generated based on the target resolution matrix and the actual resolution matrix.

[0013] Based on the first deviation matrix, the target resolution matrix is ​​iteratively adjusted to obtain the first adjusted resolution matrix;

[0014] Using the first adjusted resolution matrix as the new target resolution matrix, return to the step of dynamically enhancing resolution using the mycelial growth simulator until the structural difference is less than the preset difference threshold.

[0015] Preferably, the topological correction of the initial resolution matrix based on a preset hyphal structure feature library includes:

[0016] Identify hyphal branch nodes in the initial resolution matrix and generate a branch node distribution map;

[0017] The topological consistency coefficient is calculated based on the matching results between the branch node distribution map and the mycelial structure feature library;

[0018] When the topological consistency coefficient is lower than a preset coefficient threshold, the initial resolution matrix is ​​converted into a first feature matrix using a hyphal topological mapping model;

[0019] The first feature matrix is ​​optimized through multiple rounds using a resolution iteration engine to generate an optimized feature matrix.

[0020] The optimized feature matrix is ​​used as the new initial resolution matrix, and the step of identifying hyphal branch nodes is returned until the topological consistency coefficient reaches the preset coefficient threshold.

[0021] Preferably, the step of performing multiple rounds of optimization on the first feature matrix using a resolution iteration engine includes:

[0022] Calculate the second deviation matrix between the first feature matrix and the baseline feature matrix of the previous iteration round;

[0023] Obtain the mycelial growth state parameters for the current iteration, and generate a growth weight matrix based on the mycelial growth state parameters;

[0024] The second deviation matrix is ​​weighted using the growth weight matrix to generate the resolution adjustment amount;

[0025] The resolution adjustment is superimposed on the baseline feature matrix to generate the optimized feature matrix for the current iteration.

[0026] Preferably, obtaining the mycelial growth state parameters for the current iteration includes:

[0027] Extract the extension direction data, branch length change data, and bifurcation density change data of the hyphal branch nodes in the current iteration round;

[0028] A topological state vector is constructed based on the extension direction data, branch length variation data, and bifurcation density variation data.

[0029] The preset mycelial growth pattern library is queried using the topological state vector as an index, and the corresponding growth pattern coefficient is matched.

[0030] The growth weight matrix is ​​generated based on the growth pattern coefficients.

[0031] Preferably, after generating the first correction resolution matrix, the method further includes:

[0032] By comparing the initial resolution matrix with the first corrected resolution matrix, a resolution compensation matrix is ​​generated;

[0033] Establish a mapping model between mycelial growth temperature and the resolution compensation matrix;

[0034] The iterative adjustment of the target resolution matrix based on the first deviation matrix includes:

[0035] The corresponding resolution compensation matrix in the mapping model is invoked based on the current mycelial culture temperature.

[0036] The first deviation matrix is ​​compensated and corrected using the resolution compensation matrix to generate a third deviation matrix;

[0037] The third deviation matrix is ​​superimposed on the target resolution matrix.

[0038] Preferably, the step of returning to the dynamic resolution enhancement via the mycelial growth simulator includes:

[0039] Identify key growth regions in the first corrected resolution matrix;

[0040] Predict the required resolution level for each key growth region based on historical mycelial growth data.

[0041] Based on the resolution requirement level, the key growth region is subjected to tiered resolution enhancement.

[0042] Preferably, the resolution requirement level for predicting each key growth region based on historical mycelial growth data includes:

[0043] The frequency of occurrence of the key growth region in continuous time-series images was statistically analyzed.

[0044] Analyze the morphological complexity variation trend and location migration trajectory of the key growth regions;

[0045] The regional attention level is calculated based on the frequency of occurrence, the trend of changes in morphological complexity, and the location migration trajectory.

[0046] The regional attention level is mapped to the corresponding resolution requirement level.

[0047] Preferably, the calculation of regional attention based on the frequency of occurrence, the trend of morphological complexity change, and the location migration trajectory includes:

[0048] Divide the numerical range of regional attention and establish the correspondence between the numerical range and the resolution level;

[0049] When the actual attention given to the target critical growth area falls within a specific numerical range, the resolution level corresponding to the specific numerical range is taken as the resolution requirement level.

[0050] Preferably, generating the first deviation matrix based on the target resolution matrix and the actual resolution matrix includes:

[0051] A mycelial topology path network is constructed using the mycelial culture device as the intermediate node, different mycelial growth regions as source nodes, and the target super-resolution image as the sink node.

[0052] Based on the resolution deviation of each source node and the resolution requirement of the sink node, the capacity parameters of each path in the mycelial topology path network are calculated.

[0053] The maximum augmentable path is determined in the hyphal topology path network using the augmenting path algorithm;

[0054] The first deviation matrix is ​​generated based on the resolution transmission amount of the maximum augmentable path.

[0055] Preferably, the calculation of the capacity parameter of each path in the mycelial topology path network includes:

[0056] Obtain historical resolution stability data for each mycelial growth region;

[0057] Analyze the real-time morphological change rate of each growth region;

[0058] A dynamic reliability coefficient is generated based on the historical resolution stability data and the real-time morphological change rate.

[0059] The product of the dynamic reliability coefficient and the resolution deviation is used as the path capacity parameter.

[0060] Compared with the prior art, the beneficial effects of the present invention are:

[0061] This image super-resolution reconstruction method for mycelial topology effectively improves the quality of mycelial topology image super-resolution reconstruction through multi-stage synergy. In the initial processing stage, the initial resolution matrix of the low-resolution topology image is corrected based on a pre-set mycelial structure feature library. This can specifically identify and correct topological deviations in the low-resolution image caused by noise, imaging blur, and other issues. Since this feature library contains key feature information such as common mycelial branching angles, connection morphologies, and growth patterns, the correction process can accurately preserve the core features of mycelial topology, avoiding the neglect of specific biological structural features in traditional general correction methods, thus laying a more reliable foundation for subsequent resolution enhancement.

[0062] Dynamic resolution enhancement using a mycelial growth simulator overcomes the limitations of traditional static super-resolution methods. Mycelial growth exhibits significant dynamic characteristics, with its topological structure formation closely related to environmental stimuli and nutrient distribution during the growth process. The growth simulator can incorporate the biological laws of mycelial growth, simulating the structural evolution of mycelia under natural growth conditions during resolution enhancement. This results in an enhanced target resolution matrix that not only improves pixel details but also realistically reproduces the natural morphology of mycelial topology during growth, such as branching and adjustments in extension direction, making the reconstruction results more closely resemble the actual growth state of mycelia.

[0063] Acquiring actual topological images and extracting the actual resolution matrix using high-precision microscopic imaging equipment provides a direct reference standard for the accuracy of the reconstruction results. Comparing the target resolution matrix with the actual resolution matrix allows for the objective quantification of the structural differences between the two. When the difference exceeds a preset threshold, the target resolution matrix is ​​iteratively adjusted based on the deviation matrix, forming a closed-loop optimization mechanism. This iterative adjustment is not a simple parameter correction, but rather a targeted adjustment of deviation regions in the topological structure combined with difference analysis, such as correcting erroneous branch connections and adjusting distorted growth directions. After each iteration, the process returns to the dynamic enhancement step, allowing the growth simulator to perform dynamic optimization again based on the new target matrix. Through multiple iterations, the difference between the reconstructed result and the actual structure is gradually reduced, ultimately yielding a topological image that more closely resembles the real one.

[0064] This method organically combines topological correction, dynamic growth simulation, and iterative optimization, comprehensively considering the unique characteristics of hyphal topology and dynamic growth. Compared to traditional super-resolution methods, its reconstruction results more accurately preserve key topological features such as hyphal branching patterns, connectivity, and growth direction, reducing analytical errors caused by structural distortion. Furthermore, by referencing and iteratively adjusting actual images, the reconstruction process possesses self-optimization capabilities, adapting to differences in topological features across different growth states and species of hyphae, thus enhancing the method's applicability and reliability in processing complex hyphal samples. Attached Figure Description

[0065] Figure 1 This is a schematic diagram illustrating the working principle of the image super-resolution reconstruction method for hyphal topology described in this invention.

[0066] Figure 2 This is a flowchart of topology correction based on a mycelial structure feature library;

[0067] Figure 3 A flowchart for multi-round optimization of the resolution iteration engine;

[0068] Figure 4 This is a flowchart for resolution compensation and iterative adjustment. Detailed Implementation

[0069] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0070] Please see Figure 1This invention provides an image super-resolution reconstruction method oriented towards hyphal topology, the method comprising:

[0071] A low-resolution topological image of the mycelial sample to be processed is acquired. An initial resolution matrix characterizing its pixel information is extracted from this low-resolution image. Using a preset mycelial structure feature library, the initial resolution matrix is ​​topologically corrected to generate a first corrected resolution matrix. Subsequently, this first corrected resolution matrix is ​​input into a mycelial growth simulator. The simulator performs dynamic resolution enhancement on the matrix based on the mycelial growth kinetics and outputs a target resolution matrix. Simultaneously, an actual high-resolution topological image of the same mycelial sample is acquired using a high-precision microscopic imaging device, and the actual resolution matrix is ​​extracted from it. The structural difference between the target resolution matrix and the actual resolution matrix is ​​calculated. If the difference exceeds a preset difference threshold, a first deviation matrix is ​​calculated based on these two matrices. Based on this first deviation matrix, the previously output target resolution matrix is ​​iteratively adjusted to obtain a first adjusted resolution matrix. This first adjusted resolution matrix is ​​used as the new target resolution matrix and re-inputted into the mycelial growth simulator for a new round of dynamic resolution enhancement. This process is repeated until the calculated structural difference is lower than the preset difference threshold; the target resolution matrix at this point is the final super-resolution reconstruction result.

[0072] Example 1: See Figure 2 In the initial resolution matrix processing stage, a topology correction operation is performed. The core of this operation lies in using a pre-defined hyphal structure feature library to correct the initial data, making it more consistent with the actual morphological patterns of hyphal growth. The initial resolution matrix is ​​derived from the analysis and extraction of low-resolution hyphal topology images, and its pixel information forms the basis for subsequent processing. The identification process focuses on specific sets of pixels in the matrix that represent hyphal branching connection points. These pixels typically possess local features that distinguish them from the hyphal trunk or background, such as high gradient changes or specific neighborhood pixel patterns. By applying image processing algorithms, such as methods based on connected component analysis or morphological operations, these potential hyphal branching nodes are located. After identification, a branching node distribution map is generated. This distribution map is a data structure or image representation that clearly marks the precise coordinates of all identified branching nodes in two-dimensional or three-dimensional space, intuitively reflecting the branching topology framework of the hyphal network.

[0073] The generated branch node distribution map is matched against a pre-built hyphal structure feature library. This library is a pre-constructed knowledge base storing validated hyphal network structure feature data that conforms to biological laws. This feature data includes, but is not limited to: typical branching patterns (e.g., binary branching, lateral branching), common distance ranges between branch nodes, typical angle ranges between adjacent branches, the ratio of branch length to trunk diameter, and topological configurations specific to different hyphal species or growth stages. The matching process involves calculating the similarity between various topological indicators in the branch node distribution map (e.g., node density, nearest neighbor distance, branch angle distribution histogram) and corresponding standard models or statistical data in the feature library. Based on the matching results, a quantified topological consistency coefficient is calculated. This coefficient is a numerical indicator that comprehensively evaluates the degree of conformity between the hyphal topology represented by the initial resolution matrix and the standard structures or statistical laws in the feature library. The calculation may involve the application of distance metrics or similarity scoring functions for multidimensional feature vectors.

[0074] A preset coefficient threshold is set as the judgment standard. If the calculated topological consistency coefficient is lower than this threshold, it indicates that the hyphal topology reflected by the initial resolution matrix has a significant deviation, which may be due to noise, blurring, or imaging artifacts in the low-resolution image, leading to inaccurate branch node identification or distorted topological relationships. In this case, a correction mechanism needs to be activated. The initial resolution matrix is ​​processed using a hyphal topology mapping model. The hyphal topology mapping model is a computational model or algorithm module designed based on the basic biological principles and rules of hyphal network growth. These rules include the obstacle avoidance of hyphal growth (avoiding occupied space or unfavorable environments), chemotaxis (growing towards directions with high nutrient concentrations), the possibility of fusion between hyphae, and the triggering conditions for branch formation (such as reaching a specific length or encountering environmental stimuli). The model analyzes the initial resolution matrix, identifies potential topological connection errors, missing branches, or structures that do not conform to growth rules, and corrects or reconstructs them according to the above rules. Through the model's processing, the initial resolution matrix is ​​transformed into a new data structure, namely the first feature matrix. This first feature matrix retains the main information of the original image while expressing a topological structure that is closer to the real, biologically consistent growth morphology of hyphae.

[0075] The resolution iteration engine performs multiple rounds of optimization on the generated first feature matrix. The resolution iteration engine is an implementation module of an optimization algorithm. In each optimization iteration, the engine receives the current version of the feature matrix (initially the first feature matrix) as input. Internally, the engine applies pre-defined optimization algorithms, which may include gradient-based optimization methods (such as minimizing a defined structural loss function), heuristic search algorithms (such as simulated annealing or genetic algorithms to explore the optimal structure), or other techniques specifically designed for image structure optimization. The objective function of the optimization algorithm typically aims to minimize the difference between the current feature matrix and the ideal hyphal topology features, or to maximize its matching degree with a hyphal structure feature library. The engine attempts to improve the rationality and accuracy of the matrix in terms of topological structure by adjusting the element values ​​in the feature matrix (which may represent pixel intensity, positional offset, or structural parameters). After one round of optimization calculations, the engine outputs a new, optimized feature matrix, i.e., the optimized feature matrix.

[0076] After completing one round of optimization and generating the optimized feature matrix, this matrix is ​​not immediately used as the final output. Instead, it is fed back as a new "initial resolution matrix," re-entering the topology correction process. This means that branch node identification, generation of a new branch node distribution map, and matching with the hyphal structure feature library are performed again on this new matrix, along with a new round of topology consistency coefficient calculation. This iterative design allows the system to perform multiple, step-by-step corrections. Each cycle is based on the results of the previous optimization, further refining and correcting the hyphal topology. The cycle continues until the calculated topology consistency coefficient reaches or exceeds a preset threshold. This threshold represents an acceptable level of structural conformity. Once this condition is met, the cycle terminates. At this point, the feature matrix obtained from the last optimization, or more precisely, the matrix obtained from the last branch node identification and distribution map generation step (i.e., the matrix that meets the consistency threshold requirement), is determined as the output result of the completed topology correction, i.e., the first corrected resolution matrix. Compared to the original initial resolution matrix, this matrix has higher accuracy and biological rationality in expressing the positions and connections of hyphal branch nodes, providing structurally reliable basic data for subsequent dynamic resolution enhancement steps. The entire iterative process is automated, and the effectiveness and efficiency of the correction are ensured by using a set consistency threshold as the loop termination condition.

[0077] Example 2, see Figure 3During the resolution iteration engine's multi-round optimization of the first feature matrix, each round of optimization constitutes a complete iterative step. Its input consists of the feature matrix to be optimized in the current round (the initial input is the first feature matrix) and the baseline feature matrix generated in the previous round (for the first round, the baseline feature matrix can be set as the initial input or a zero matrix). The output is the optimized feature matrix adjusted in this round. The core of this optimization process lies in dynamically adjusting the optimization direction and intensity of the matrix based on the simulated real-time mycelial growth state.

[0078] At the start of each optimization round, the difference between the current input feature matrix (i.e., the first feature matrix to be optimized in this round) and the baseline feature matrix obtained in the previous round is calculated. This difference is achieved through element-wise comparison or other matrix differencing operations, ultimately generating a second deviation matrix that reflects the deviation of pixel values ​​or structural parameters between the two. This matrix quantifies the amount of change in the matrix state since the previous optimization round.

[0079] The system needs to acquire state parameters characterizing the simulated hyphal growth dynamics in the current iteration. Acquiring these parameters involves a data extraction and analysis process. Specifically, it analyzes the hyphal structure data represented by the currently input feature matrix, extracting the spatial extension direction information of each identified hyphal branch node. This typically involves calculating the direction vector between a node and its neighboring nodes or growth tips, and may involve normalization or statistical analysis of the principal direction. The system also calculates the change in distance between adjacent branch nodes. This is achieved by comparing the node spacing in the current matrix with historical records (such as the spacing in the previous round's baseline matrix) or a preset standard growth step size, obtaining branch length variation data that reflects the rate of hyphal extension or contraction. Furthermore, it statistically analyzes the change in the number of newly formed branch points within a specific unit area defined by the feature matrix or along a specific hyphal trunk unit length. This requires identifying newly added branch nodes in the current round and comparing them with previous states to obtain bifurcation density variation data, indicating the activity level of hyphal branch formation.

[0080] The extracted data on extension direction, branch length variation, and bifurcation density variation are integrated to construct a multidimensional topological state vector. Each dimension of this vector corresponds to a specific growth state index, whose values ​​comprehensively describe the overall growth status and local dynamics of the mycelial network at the current simulation moment. For example, one dimension might represent the sine value of the average extension direction angle, another dimension might represent the average rate of change of branch length, and a third dimension might represent the increment of bifurcation density per unit area.

[0081] The constructed topological state vector is used as a query index to access a pre-defined mycelial growth pattern library. This growth pattern library is a pre-generated database or knowledge base that stores combinations of mycelial growth parameters observed or simulated at different typical growth stages (such as the rapid linear extension phase, the branching burst phase, the slow growth phase caused by nutrient absorption, and the environmental stress response phase), along with their corresponding pattern coefficients. These pattern coefficients are essentially a set of weighting factors that define the relative importance of different spatial locations, structural features, or growth behaviors in overall optimization under specific growth states. Through vector similarity calculations (such as cosine similarity, Euclidean distance comparison, etc.) or pattern matching algorithms, the system searches the growth pattern library for the entry most similar to or best matching the currently input topological state vector and extracts the growth pattern coefficients associated with that entry.

[0082] Based on the successfully matched growth pattern coefficients, the system generates a growth weight matrix. The dimensional structure of this matrix is ​​consistent with the second deviation matrix. The value of each element in the growth weight matrix is ​​not fixed but dynamically determined by the acquired growth pattern coefficients. Its assignment logic reflects the influence weight of the current simulated growth state on different parts of the mycelial network. For example, if the current state is identified as a rapid linear extension phase, pixels or structural parameters located in the leading edge region of active growth tips may be assigned higher weights; if it is in a branching burst phase, the weights of regions near historical branch nodes or in the direction of specific environmental stimuli may increase; and during the nutrient absorption phase, the weights of regions near nutrient sources may increase. The value range of the weight matrix is ​​typically in the [0,1] interval or after normalization; a larger value indicates that the location or feature needs more significant adjustment in this optimization iteration.

[0083] After generating the growth weight matrix, it is applied to the previously calculated second deviation matrix. Specifically, this is achieved by performing element-wise multiplication or other forms of weighted operations on the two matrices. The result of this operation is a new matrix called the resolution adjustment matrix. The resolution adjustment matrix inherits the adjustment direction indicated by the second deviation matrix (positive or negative sign represents increase or decrease), but its adjustment magnitude (absolute value) is scaled according to the growth weight matrix. Adjustments at high-weight positions are amplified, while adjustments at low-weight positions are reduced or even suppressed. In this way, the resolution adjustment matrix dynamically incorporates information about the current simulated mycelial growth state, making the optimization process more closely aligned with the dynamic laws of biology.

[0084] Finally, the generated resolution adjustment matrix is ​​superimposed on the baseline feature matrix obtained from the previous optimization round. The superposition operation is typically an element-wise addition of matrices. The result of the superposition is a new matrix, the optimized feature matrix for the current iteration. This optimized feature matrix integrates the state from the previous round and the adjustment dynamically calculated based on the growth state in this round, representing the new state of the hyphal structure after this optimization. This optimized feature matrix will serve as the baseline feature matrix for the next optimization iteration (used to calculate the second deviation matrix for the next round), or, if the iteration termination condition is met (such as reaching a preset number of rounds or optimization convergence), it will serve as the final output of the resolution iteration engine for this round. Through this weighted adjustment mechanism combining real-time growth state parameters, the resolution iteration engine can more intelligently and biologically conform to multiple rounds of progressive optimization of the first feature matrix, gradually improving its accuracy and rationality in representing the hyphal topology. The entire optimization loop is closed and self-consistent; each round depends on the result of the previous round and generates the input for the next round, until the optimization goal is achieved.

[0085] Example 3: See Figure 4 After obtaining the first corrected resolution matrix, additional processing steps are performed to enhance the system's adaptability to environmental factors and the specificity of subsequent processing. First, the original initial resolution matrix is ​​compared with the first corrected resolution matrix generated after topology correction. This comparison typically involves element-wise difference calculations or other difference measurement methods to quantify the information changes introduced by the topology correction process. This comparison generates a resolution compensation matrix. This matrix has the same dimensions as the original matrix, and each element records the change in intensity or eigenvalue of the corresponding spatial pixel due to correction; positive values ​​may indicate enhancement, while negative values ​​may indicate suppression or correction.

[0086] A quantitative correlation model was established between the environmental temperature parameters of mycelial culture and the resolution compensation matrix. This mapping model is based on the understanding of the temperature effect on mycelial growth. By culturing mycelial samples under different set temperature conditions and simultaneously acquiring their topological images, the influence of temperature changes on mycelial morphological characteristics (such as branching angle, growth rate, and density distribution) was analyzed. These observed influence patterns were quantified and transformed into correction factors or offsets for the elements of the resolution compensation matrix. Ultimately, the model is represented as a function or lookup table, with the current temperature value as input and the corresponding resolution compensation matrix reflecting typical morphological change patterns at that temperature as output, denoted as [missing information - likely a typo]. ,in This represents the real-time monitoring or set temperature value for mycelial culture.

[0087] In the subsequent core step of iteratively adjusting the target resolution matrix based on the first deviation matrix, the aforementioned temperature compensation mechanism is introduced. The specific adjustment process is as follows: real-time acquisition of the temperature value of the culture environment where the current mycelial sample is located. Based on this temperature value Query and call the previously established mapping model to obtain the corresponding resolution compensation matrix. The purpose of calling this matrix is ​​to incorporate the potential impact of temperature on hyphal morphology into the current bias correction process. Using the obtained... For the first deviation matrix Compensation and correction operations are performed. The specific operation can be matrix addition, element-wise weighted combination, or other linear / nonlinear operations, aiming to superimpose or integrate the temperature effect onto the original deviation information. This operation produces a new matrix, called the third deviation matrix. Its mathematical expression can be represented as:

[0088]

[0089] in: This represents a specific compensation function, such as simple addition. , This can be an adjustable compensation intensity coefficient, or a more complex rule-based fusion. Finally, this calculated third deviation matrix... Superimposed (usually matrix addition) onto the target resolution matrix to be adjusted This completes the current iterative adjustment, resulting in the first adjusted resolution matrix. ,Right now .

[0090] After completing the iterative adjustments described above and generating the first adjusted resolution matrix, this matrix will be used as new input to return to the step of dynamic resolution enhancement via the mycelial growth simulator. Before re-executing this enhancement step, a process of analysis and enhancement strategy formulation for key growth regions is also included. Specifically, the matrix to be input into the simulator (currently the first adjusted resolution matrix or the previous calibration matrix) is analyzed to identify specific regions that significantly influence the overall mycelial growth dynamics; these regions are defined as key growth regions. Identification criteria may include: the region's growth activity (e.g., tip growth rate), structural complexity (e.g., branch density), the significance of dynamic changes shown in historical images, or its location at a potential interaction point (e.g., mycelial confluence area). The identification algorithm may be based on edge detection, region segmentation, activity analysis, or location filtering combined with prior knowledge.

[0091] After identifying key growth regions, based on historical mycelial growth time-series image datasets, the required resolution level (resolution requirement level) for each key growth region at the current simulation time point or growth stage is predicted. The prediction process utilizes image evolution patterns of similar growth stages and region types in historical data. For example, analyzing the morphological detail complexity of active growth tip regions in historical sequences at corresponding periods infers the required high resolution level for the current tip; or analyzing image feature changes in branching junctions before fusion to predict their current high resolution requirement. The prediction result assigns a resolution requirement level to each key growth region, which can be discrete (e.g., high, medium, low) or a continuous numerical scale.

[0092] Based on the predicted resolution requirement levels of each key growth region, a hierarchical resolution enhancement strategy is implemented during subsequent dynamic resolution enhancement using a mycelial growth simulator. The core of this strategy is applying enhancement processing of different intensities or algorithms to regions with different requirement levels. For example, for key regions predicted to have a high resolution requirement level (such as active growth tips), the simulator may apply a more refined, computationally intensive super-resolution algorithm model, or perform a higher resolution enhancement in that region to capture subtle morphological changes and edge details. For medium-level regions (such as stably extending trunks), standard-intensity enhancement algorithms are applied. For low-level regions (such as background or dormant areas), more basic or faster low-intensity enhancement processing may be applied, or even the original resolution may be maintained to reduce computational overhead. This hierarchical approach ensures that limited computational resources are prioritized for regions most critical to understanding mycelial growth dynamics, optimizing overall processing efficiency and the biological relevance of the results. The entire process, from temperature compensation to hierarchical enhancement of key regions, constitutes a dynamic resolution enhancement scheme that is environmentally sensitive and focused on growth hotspots.

[0093] Example 4: In the process of predicting the resolution requirement level of each key growth region based on historical mycelial growth data, it is necessary to process continuously acquired time-series image datasets and perform dynamic analysis on the key growth regions identified in the current image. Assume that in a certain processing step, the system identifies three key growth regions from the current first corrected resolution matrix, labeled as region A (active growth tip), region B (dense branching area), and region C (potential fusion point). The specific operation for predicting the resolution requirement level of these regions is as follows.

[0094] Access the stored historical hyphal growth time-series image sequence. This sequence consists of high-resolution images periodically acquired from similar hyphal samples under identical culture conditions, spanning multiple growth stages. For each key growth region in the current image, perform three data analysis tasks. The first task is to count the frequency of this region's occurrence in consecutive historical images. This requires defining identifiability criteria for the region (e.g., minimum area, specific morphological features) and retrospectively searching for region instances that meet these criteria in the historical image sequence. Count the number of times it appears in the most recent N frames (e.g., the most recent 10 frames) as the frequency data. A higher frequency indicates that the region persists during growth and is likely important.

[0095] The second task is to analyze the trend of morphological complexity variation in this key growth region. Morphological complexity is a quantitative indicator that reflects the degree of irregularity of the region's boundary or the fineness of its internal structure. Calculation methods include, but are not limited to: extracting the region's contour coordinates and calculating its fractal dimension; or performing texture analysis on the region's interior and calculating the contrast or entropy value of the gray-level co-occurrence matrix. In historical image sequences, the morphological complexity value of this region (or its most similar region) is tracked over time. Through linear fitting or curve analysis, the trend is determined to be increasing (complexity rising), decreasing (complexity falling), or relatively stable. For example, a region forming a rapid branch may exhibit a significant increasing trend in morphological complexity.

[0096] The third task is to track the location migration trajectory of this key growth region. In continuous historical images, the position sequence of this region (or its corresponding region) in the image coordinate system is determined using feature matching or centroid tracking algorithms. Its movement trajectory is analyzed to calculate the average migration speed, frequency of direction changes, or trajectory curvature. For example, a growth tip migrating directionally towards a nutrient source will exhibit a higher linear migration speed, while regions hovering near obstacles will show frequent changes in direction.

[0097] After completing the above three analyses, a comprehensive regional attention value is calculated for each key growth region. This calculation uses a weighted summation model to integrate the quantitative results of three dimensions: frequency of occurrence (F), morphological complexity trend (T), and location migration trajectory (P). The quantitative values ​​of each dimension need to be normalized to the [0,1] interval. The frequency of occurrence F is directly calculated by dividing the number of occurrences by the total number of frames. The morphological complexity trend T is assigned a value based on the trend type: a significantly increasing trend is assigned a higher value (e.g., 0.8), a stable trend is assigned a medium value (e.g., 0.5), and a decreasing trend is assigned a lower value (e.g., 0.2). The location migration trajectory P is scored based on a comprehensive evaluation of migration speed and directional stability: high-speed directional migration is assigned a high score (e.g., 0.9), and low-speed irregular movement is assigned a low score (e.g., 0.3). Weight coefficients are set for each dimension, for example, the weight of frequency of occurrence W_f = 0.3, the weight of morphological complexity trend W_t = 0.4, and the weight of migration trajectory W_p = 0.3. The formula for calculating regional attention is: Attention = (W_f * F) + (W_t * T) + (W_p * P). The result is a value between 0 and 1.

[0098] Table 1: Calculated data for the three identified key growth regions (A, B, C).

[0099]

[0100] After calculating the regional attention value, it is mapped to the corresponding resolution requirement level. The mapping rule is based on a preset range of attention values. The system defines three ranges: [0, 0.3) corresponds to low resolution requirement level, (0.3, 0.7) corresponds to medium resolution requirement level, and [0.7, 1.0] corresponds to high resolution requirement level. According to the calculation results in the table above: the attention value of region A is 0.87, which falls into the range [0.7, 1.0], so its resolution requirement level is mapped to "high"; the attention value of region B is 0.57, which falls into the range (0.3, 0.7), and is mapped to "medium"; the attention value of region C is 0.29, which falls into the range [0, 0.3), and is mapped to "low".

[0101] This prediction directly guides subsequent hierarchical resolution enhancement processing. For example, for region A (high demand), the system may allocate more computational resources, employ more complex super-resolution algorithms (such as deep learning-based models), and perform higher upsampling to finely reconstruct the delicate structures of its active growth tips. For region B (medium demand), standard super-resolution algorithms (such as interpolation combined with edge enhancement) are used for moderate resolution enhancement. For region C (low demand), only basic bilinear interpolation may be applied or the current resolution may be maintained to save overall computational overhead. This approach, based on historical dynamic data prediction and hierarchical processing, aims to optimize resource allocation, enabling high-interest key growth regions to achieve more refined imaging results. The entire process relies on quantitative analysis and rule mapping of historical growth patterns, without pre-setting fixed regional importance.

[0102] Example 5: In the process of generating the first deviation matrix, an optimization model based on network flow theory is used to allocate resolution enhancement resources. This process begins with obtaining the target resolution matrix and the actual resolution matrix. The target resolution matrix is ​​the matrix output after dynamic enhancement by the mycelial growth simulator, representing the predicted high-resolution structure. The actual resolution matrix comes from the direct acquisition and extraction of the same mycelial sample by a high-precision microscopic imaging device, representing real, high-precision mycelial topology information. Calculating the difference between the two is the basis for generating the first deviation matrix, but this example introduces a network flow model to intelligently allocate adjustment amounts.

[0103] A virtual path network reflecting the mycelial topology and processing relationships is constructed. This network contains three types of nodes: source nodes, intermediate nodes, and sink nodes. Source nodes correspond to different growth regions divided within the mycelial network. These regions can be divided based on spatial location (e.g., dividing the mycelial image into grids), functional characteristics (e.g., active growth regions, dormant regions, branching node regions), or dynamic importance (e.g., key growth regions identified in the aforementioned embodiments). Each source node is associated with the difference between the target resolution matrix and the actual resolution matrix of a local region, i.e., the resolution deviation of that region. Intermediate nodes are defined as mycelial culture devices. Here, the culture device does not refer to the entire physical equipment, but rather to a specific location point or environmental control unit in the culture environment that has a regulatory or influencing effect, such as the central nutrient source location of the petri dish, a specific temperature control zone, or the location of a humidity sensor. Setting the culture device as an intermediate node aims to reflect the mediating role of environmental factors in the resolution information transmission path. Sink nodes are defined as the overall target super-resolution image to be output. Directed connection paths are established between source nodes, intermediate nodes, and sink nodes. The rules for establishing paths simulate the actual growth paths or logical connections of hyphae: for example, a path from a source node (a specific growth region) to an intermediate node (such as the location of a nutrient source) may represent the intensity of the influence of the environmental factor on that region; a path from an intermediate node to a sink node represents the contribution channel of the environmental factor to the final image synthesis; in some cases, the source node may also directly point to the sink node according to the topological connection, representing the direct influence of the information of that region on the final image.

[0104] The capacity parameter of each directed path in the mycelial topology network is calculated. The capacity parameter characterizes the upper limit of the resolution information that the path can transmit. Its calculation depends not only on the resolution deviation of the source nodes (i.e., the magnitude of enhancement required for the region) but also on the overall resolution requirement of the sink nodes (i.e., the desired sharpness level of the final image) and is constrained by the reliability of the path itself. The specific calculation process involves three data inputs: First, historical resolution stability data for each source node (i.e., each mycelial growth region) is obtained. This is obtained by analyzing the fluctuations in the region's resolution performance in continuous historical time-series images, such as calculating the variance, standard deviation, or maximum and minimum difference of its pixel values ​​or feature values ​​within a certain time window. Historical stability data reflects the reliability of the imaging results for that region; high stability means more reliable deviations. Second, the morphological change rate of each growth region in the current real-time imaging frame is analyzed. This involves calculating the movement speed of the boundary pixels of the region, the instantaneous change rate of internal texture features, or the expansion / contraction rate of the region's area. The real-time morphological change rate indicates the dynamic activity of the region; regions with high activity may be more unstable in their current state. Based on the historical stability data and real-time morphological change rate, a dynamic reliability coefficient is generated using a pre-defined algorithm model. This coefficient is a value between 0 and 1, and its calculation may assign higher weight to historical stability (because it is based on more data), while also considering the real-time change rate; an excessively high change rate may temporarily reduce reliability. The dynamic reliability coefficient quantifies the current reliability of the source node's resolution deviation. Finally, the capacity parameter of the path is determined as: the absolute value of the source node's resolution deviation multiplied by the calculated dynamic reliability coefficient. In this way, the capacity parameter considers both the magnitude of the deviation and the reliability of the deviation information.

[0105] After calculating the capacity parameters for all paths, the augmenting path algorithm is applied to find the optimal combination of augmenting paths in the mycelial topology path network. The augmenting path algorithm is a classic algorithm in network flow theory used to solve the maximum flow problem (such as the Ford-Fulkerson algorithm). The goal of the algorithm is to find a set of paths from the source nodes through intermediate nodes to the sink node in the network, maximizing the total resolution information transmitted from all source nodes to the sink node without exceeding the capacity limit of each path. The algorithm execution process involves iterative search: starting from zero flow, it continuously searches for paths (augmenting paths) in the residual network that can increase flow, and increases flow along these paths until no new augmenting paths can be found. The final determined maximum augmentable path set represents the maximum total resolution enhancement that can be transmitted to the sink node (target image) under path capacity constraints, and specifies which source nodes (growth regions) these enhancements originate from and through which paths (environmental mediation or direct connections).

[0106] Based on the found maximum augmentable path set and the resolution information (i.e., path flow) transmitted by each path, a first deviation matrix is ​​generated. This matrix has the same dimension as the target / actual resolution matrix. The allocation logic of its elements is determined by the network flow optimization results: for each image location or region, the corresponding deviation adjustment depends on the actual flow used by all paths originating from that location / region in the maximum flow scheme. Higher flow means a greater resolution enhancement allocated to that location / region. More importantly, the optimization process prioritizes high-confidence source nodes (i.e., regions with high dynamic confidence coefficients) to receive higher transmission priority and greater flow allocation, while also considering the overall resolution requirements of sink nodes. The final generated first deviation matrix not only reflects the difference between the target and actual matrices but also incorporates intelligent adjustment strategies based on network topology, historical stability, real-time dynamics, and environmental mediation, providing a highly directional and reliable correction basis for subsequent iterative adjustments. The entire network flow model construction, capacity calculation, and augmenting path search constitute a structured and optimized deviation information extraction and allocation mechanism.

[0107] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0108] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for image super-resolution reconstruction oriented to mycelium topology, characterized in that, The method comprises the following steps: acquiring a low-resolution topological image of a mycelium sample, and extracting an initial resolution matrix of the low-resolution topological image; performing topological structure correction on the initial resolution matrix based on a preset mycelium structure feature library to generate a first corrected resolution matrix; performing dynamic resolution enhancement on the first corrected resolution matrix by a mycelium growth simulator to output a target resolution matrix; acquiring an actual topological image of the mycelium sample by using a high-precision microscopic imaging device, and extracting an actual resolution matrix from the actual topological image; when a structural difference degree between the target resolution matrix and the actual resolution matrix exceeds a preset difference threshold, generating a first deviation matrix according to the target resolution matrix and the actual resolution matrix; performing iterative adjustment on the target resolution matrix based on the first deviation matrix to obtain a first adjusted resolution matrix; taking the first adjusted resolution matrix as a new target resolution matrix, and returning to the step of performing dynamic resolution enhancement by the mycelium growth simulator until the structural difference degree is less than the preset difference threshold.

2. The mycelium topology oriented image super-resolution reconstruction method according to claim 1, characterized in that, The step of performing topological structure correction on the initial resolution matrix based on the preset mycelium structure feature library comprises the following steps: identifying mycelium branch nodes in the initial resolution matrix to generate a branch node distribution map; calculating a topological consistency coefficient according to a matching result of the branch node distribution map and the mycelium structure feature library; when the topological consistency coefficient is lower than a preset coefficient threshold, converting the initial resolution matrix into a first feature matrix by using a mycelium topological mapping model; performing multi-round optimization on the first feature matrix by a resolution iteration engine to generate an optimized feature matrix; taking the optimized feature matrix as a new initial resolution matrix, and returning to the step of identifying mycelium branch nodes until the topological consistency coefficient reaches the preset coefficient threshold.

3. The mycelium topology oriented image super-resolution reconstruction method according to claim 2, characterized in that, The step of performing multi-round optimization on the first feature matrix by the resolution iteration engine comprises the following steps: calculating a second deviation matrix of the first feature matrix and a reference feature matrix of a previous iteration round; acquiring mycelium growth state parameters of a current iteration round, and generating a growth weight matrix based on the mycelium growth state parameters; performing weighted processing on the second deviation matrix by using the growth weight matrix to generate a resolution adjustment amount; adding the resolution adjustment amount to the reference feature matrix to generate an optimized feature matrix of the current iteration round.

4. The mycelium topology oriented image super-resolution reconstruction method according to claim 3, characterized in that, The step of acquiring mycelium growth state parameters of the current iteration round comprises the following steps: extracting extension direction data, branch length change data and bifurcation density change data of mycelium branch nodes in the current iteration round; constructing a topological state vector according to the extension direction data, the branch length change data and the bifurcation density change data; querying a preset mycelium growth mode library by using the topological state vector as an index to match corresponding growth mode coefficients; generating the growth weight matrix based on the growth mode coefficients.

5. The mycelium topology oriented image super-resolution reconstruction method according to claim 1, characterized in that, After the first corrected resolution matrix is generated, the method further comprises the following steps: comparing the initial resolution matrix and the first corrected resolution matrix to generate a resolution compensation matrix; establishing a mapping relationship model between mycelium growth temperature and the resolution compensation matrix; the iterative adjustment of the target resolution matrix based on the first deviation matrix includes: calling the corresponding resolution compensation matrix in the mapping relationship model according to the current mycelium culture temperature; compensating and correcting the first deviation matrix by using the resolution compensation matrix to generate a third deviation matrix; superimposing the third deviation matrix on the target resolution matrix.

6. The mycelium topology oriented image super-resolution reconstruction method according to claim 1, characterized in that, the step of returning the dynamic resolution enhancement through the mycelium growth simulator includes: identifying the key growth area in the first corrected resolution matrix; predicting the resolution requirement level of each key growth area according to historical mycelium growth data; based on the resolution requirement level, the key growth area is subjected to hierarchical resolution enhancement.

7. The mycelium topology oriented image super-resolution reconstruction method according to claim 6, characterized in that, the resolution requirement level of each key growth area is predicted according to historical mycelium growth data, which includes: statistical analysis of the frequency of occurrence of the key growth area in the continuous time sequence image; analyze the morphological complexity change trend and position migration trajectory of the key growth area; calculate the area attention according to the frequency of occurrence, morphological complexity change trend and position migration trajectory; map the area attention to the corresponding resolution requirement level.

8. The mycelium topology oriented image super-resolution reconstruction method according to claim 7, characterized in that, the area attention is calculated according to the frequency of occurrence, morphological complexity change trend and position migration trajectory, which includes: dividing the numerical interval of the area attention, and establishing the corresponding relationship between the numerical interval and the resolution level; when the actual attention of the target key growth area falls into a specific numerical interval, the resolution level corresponding to the specific numerical interval is taken as the resolution requirement level.

9. The mycelium topology oriented image super-resolution reconstruction method according to claim 1, characterized in that, the first deviation matrix is generated according to the target resolution matrix and the actual resolution matrix, which includes: taking mycelium culture device as intermediate node, different mycelium growth area as source node, and target super-resolution image as sink node, a mycelium topology path network is constructed; according to the resolution deviation of each source node and the resolution requirement value of the sink node, the capacity parameter of each path in the mycelium topology path network is calculated; determine the maximum enhancement path in the mycelium topology path network by the augmented path algorithm; based on the resolution transmission capacity of the maximum enhancement path, the first deviation matrix is generated.

10. The mycelium topology oriented image super-resolution reconstruction method according to claim 9, characterized in that, the capacity parameter of each path in the mycelium topology path network is calculated, which includes: obtain the historical resolution stability data of each mycelium growth area; analyze the real-time morphological change rate of each growth area; generate a dynamic credibility coefficient according to the historical resolution stability data and real-time morphological change rate; the product of the dynamic credibility coefficient and the resolution deviation is taken as the path capacity parameter.

Images