Method for analyzing spore propagation of aspergillus oryzae in koji-making process based on microscopic image recognition
Patent Information
- Application Number
- CN202610994366.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-07-06
- Publication Date
- 2026-09-25
AI Technical Summary
[0003]当前市面主流的制曲米曲霉孢子检测方法高度依赖操作人员经验，采样制片、显微观测无标准化流程，成像参数不统一，人为主观性误差大、重复性差，且检测效率低下，无法实现高通量批量检测
实现从样本采集制备、显微图像采集到数据分析的全流程标准化、自动化作业，通过统一采样位置、成像光照与曝光参数，规避传统人工操作的随机性与设备参数差异带来的检测误差。依托多维度图像预处理、自适应阈值结合边缘检测的精准分割技术，搭配先验特征过滤与机器学习分类器的双重识别机制，可高效剔除曲料杂质干扰，精准区分米曲霉孢子与干扰物，大幅提升孢子识别的准确率与稳定性。同时该方法能够高通量获取孢子浓度、形态分布等多维数据，通过时序数据拟合生成高精度繁殖动力学与生长曲线，可精准划分孢子繁殖阶段、量化繁殖速率，弥补了传统方法效率低、主观性强、无法精细化表征孢子动态繁殖规律的缺陷，为制曲过程微生物生长调控提供精准、客观、连续的数据支撑。
Smart Images
Figure CN122821550A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of image processing, and in particular to a method for analyzing the reproduction of Aspergillus oryzae spores during the koji-making process based on microscopic image recognition. Background Technology
[0002] Koji making is the core process in the production of traditional fermented foods such as soy sauce, rice wine, and broad bean paste. As a dominant functional strain, Aspergillus oryzae's spore germination and proliferation directly determine the secretion level of enzyme system in the koji material, fermentation stability, and the flavor quality of the finished product. Accurately tracking the spore reproduction dynamics is a key prerequisite for precise control of the koji making process.
[0003] Current mainstream methods for detecting Aspergillus oryzae spores in koji-making processes heavily rely on operator experience. Sampling, slide preparation, and microscopic observation lack standardized procedures, resulting in inconsistent imaging parameters, significant subjective errors, poor repeatability, and low detection efficiency, making high-throughput batch detection impossible. Traditional image detection techniques employ simple preprocessing methods, often using fixed threshold segmentation, which are unsuitable for complex backgrounds and uneven lighting conditions in koji-making processes. They also have weak resistance to interference from impurities, easily leading to missed or false spore detections. Furthermore, existing methods can only perform simple spore counts, lacking multi-dimensional feature screening and intelligent model-based fine discrimination, resulting in limited recognition accuracy. They also only acquire scattered data at discrete time points, failing to generate continuous reproductive dynamics and growth curves, making it difficult to accurately classify spore reproduction stages, quantify reproduction rates, and precisely and dynamically characterize the reproductive patterns of Aspergillus oryzae, thus failing to meet the precise monitoring and scientific control requirements of modern koji-making processes. Summary of the Invention
[0004] To improve existing methods, this paper provides a method for analyzing the reproduction of Aspergillus oryzae spores during the koji-making process based on microscopic image recognition. This method relies on standardized microscopic imaging, machine vision, and machine learning technologies. It has strong anti-interference capabilities, high spore recognition accuracy, and can acquire time-series reproduction data in high throughput. It can accurately fit growth kinetic curves, quantify reproduction rates, and continuously and meticulously characterize the reproduction dynamics of Aspergillus oryzae spores during the koji-making process.
[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A method for analyzing the reproduction of Aspergillus oryzae spores during the koji-making process based on microscopic image recognition includes: At key stages of koji making, samples are collected from different locations of the koji material, and after low-temperature transfer and aseptic grinding, a uniform spore suspension is prepared. The suspension is allowed to stand and the supernatant is taken and quantitatively added to a standard counting plate to prepare a standardized sample with uniform thickness. The standardized sample is placed on the microscope stage, and multiple field-of-view positions are planned according to the preset grid. After focusing at each position, high-definition digital microscopic images are acquired under constant illumination and exposure parameters to form an image sequence covering the sample area. The image sequence was preprocessed with white balance correction, noise filtering and illumination homogenization. Contrast was enhanced by converting the color space. The image was binarized and segmented using an adaptive thresholding method combined with edge detection. Candidate spore connected regions were preliminarily screened through morphological operations. Based on the selected spore connectivity regions, the morphological, color, and texture features of each region are calculated. Preliminary filtering is performed based on the prior feature range of Aspergillus oryzae spores to remove obvious impurities. A pre-trained machine learning classifier is used to finely discriminate the remaining targets, identify Aspergillus oryzae spores, and generate contour masks. Based on the identification results, the spore concentration per unit volume is statistically analyzed, and the morphological distribution of the population spores is analyzed. The spore parameters at different time points are arranged in a time series to generate reproductive kinetic curves and growth curves, identify the reproductive stages, and calculate the reproductive rate of specific stages.
[0006] Preferably, the image sequence forming the sample area specifically includes: Place the prepared standardized sample on the microscope stage; Within the sample observation area, multiple non-overlapping field-of-view locations are planned according to a pre-defined grid matrix; At each preset field-of-view location, the microscope autofocus system executes a focusing algorithm based on image contrast to acquire the optimal focal plane image; Under constant light intensity and uniform exposure parameters, the camera sequentially acquires raw RGB color microscopic digital images of each field of view, forming a high-throughput image sequence covering the sample area; Record metadata for each image, including acquisition time, field of view coordinates, objective magnification, and light source parameters.
[0007] Preferably, the screening of candidate spore connectivity regions specifically includes: The acquired high-throughput raw microscopic image sequences were standardized and preprocessed, and all raw images were uniformly white-balanced to eliminate color temperature differences. A Gaussian filtering-based method is used for noise suppression, which smooths the image background while preserving spore edge details. To address potential uneven lighting, a background subtraction algorithm is applied to obtain an image with uniform brightness. The preprocessed RGB image is converted to a color space to enhance the contrast between the spores and the background and impurities; An adaptive threshold segmentation algorithm, combined with the assistance of an edge detection operator, is used to dynamically determine the optimal segmentation threshold for each image or local region of an image, thereby generating a binarized image. Morphological operations are performed on the binarized image. Opening operations are used to remove small noise points, and closing operations are used to fill the cavities inside the spores and smooth the spore outlines, thus initially obtaining the connected regions of candidate spores.
[0008] Preferably, the specific process for generating the binarized image includes: The preprocessed grayscale image is divided into multiple overlapping or adjacent local regions. For each local region, the weighted average of the pixel grayscale values within the region is calculated as the initial local threshold for the center pixel position. An initial threshold map is generated by traversing the entire image. The edge detection operator is used to scan the image to identify edge pixels where the gray level changes abruptly at the boundary between the spore and the background. The thresholds at the corresponding positions in the initial threshold map are then fine-tuned and corrected to obtain an optimized fine threshold map. For each pixel in the original image, the dynamic segmentation threshold for that pixel location is read from the fine threshold map based on its coordinates. The gray value of the pixel is compared with the dynamic threshold to divide the foreground and background, generating a preliminary binarized image. Aspergillus oryzae spores appear as white connected regions, while the background is black.
[0009] Preferably, the step of identifying Aspergillus oryzae spores and generating a contour mask includes: For each candidate spore connected region obtained, its morphological, color, and texture features are calculated, and a feature vector is constructed to describe it; The morphological features include region area, perimeter, equivalent diameter, aspect ratio, roundness, and convexity; the color features include the average intensity and intensity standard deviation of the connected regions in the original RGB or converted color space; and the texture features are obtained by calculating the local binary pattern. A feature filter based on prior knowledge is set up to initially screen candidate targets according to the typical feature range of pure Aspergillus oryzae spores under the microscopic conditions used, and obvious non-spore targets with irregular size, shape and color deviation are eliminated. For targets that pass the initial screening, a classifier based on a pre-trained machine learning model is used for fine classification. The model is trained using labeled positive samples of Aspergillus oryzae spores and negative samples of non-spores. The final output is a target mask image that is confirmed to be Aspergillus oryzae spores.
[0010] Preferably, the feature filter based on prior knowledge, which preliminarily screens candidate targets according to the typical characteristic range of pure Aspergillus oryzae spores under the applied microscopic conditions, and eliminates obvious non-spore targets with irregular size, shape, or color deviation, specifically includes: Under the same microscopic imaging conditions, images of known pure Aspergillus oryzae spores were pre-acquired and analyzed, the distribution range of the population on key features was statistically analyzed, and a prior knowledge base was constructed. Based on the constructed prior knowledge base, a series of logical judgment conditions are set to eliminate candidate targets whose equivalent diameter is smaller or larger than the normal range of spores, eliminate candidate targets whose roundness is lower than the threshold or whose length-to-short axis ratio is higher than the threshold, and exclude candidate targets that deviate from the typical range of spores in terms of key color features. The feature vector of each candidate target is compared with the multi-level filtering rules in turn, and the candidate targets that pass all logical judgment conditions are retained.
[0011] Preferably, the classifier of the pre-trained machine learning model specifically includes: Annotated image patches are extracted from historical microscopic images to construct a training dataset. Positive samples are image patches containing complete and typical Aspergillus oryzae spores, while negative samples contain interference objects with similar morphology or color to spores. For each training sample image patch, a multi-dimensional feature vector containing all morphological, color, and texture features is generated. The model is trained by inputting multi-dimensional feature vectors using a machine learning algorithm based on support vector machines. After training, the model performance is evaluated using a validation set. For candidate targets that pass the initial screening, extract the image patch of the region where it is located, and calculate the same multi-dimensional feature vector and input it into the pre-trained model to calculate the probability that the target belongs to the Aspergillus oryzae spore category, and determine whether the target is an Aspergillus oryzae spore; For all targets identified as spores, the corresponding connected regions in their binarized mask images are marked as valid spore regions, ultimately generating a target mask image containing only the identified spores.
[0012] Preferably, the generation of reproductive kinetic curves and growth curves, identification of reproductive stages, and calculation of the reproductive rate at specific stages specifically includes: Based on the Aspergillus oryzae spore targets confirmed by fine screening, the number of spores identified in each microscopic image was calculated; By combining the microscope objective magnification, camera pixel size, and known volume depth of the counting chamber, the number of spores per field of view is converted into the spore concentration per unit volume of sample solution; The spore concentrations of all effective fields of view of the current sample are statistically analyzed, and their average value and standard deviation are calculated as the Aspergillus oryzae spore density at this koji-making time point. Analyze the distribution of morphological characteristics of all identified spores; The parameters obtained at different music production time points are arranged in a time series. The spline interpolation method was used to generate the reproductive kinetic curve of Aspergillus oryzae spore density over time and the growth curve of average spore size over time during the koji-making process. Based on the reproductive kinetics curve and growth curve, the spore germination period, rapid reproduction period and stationary period are identified, and the reproduction rate of specific stages is calculated.
[0013] Preferably, the generation of the reproductive kinetic curve of Aspergillus oryzae spore density over time and the growth curve of average spore size over time using spline interpolation specifically includes: The measured data obtained through microscopic image analysis at different time points in the koji-making process were collected to form a time-spore density dataset and a time-spore average diameter dataset. The interval between adjacent measured time points is defined as a segment, and a smooth mathematical curve is constructed independently for each segment. The segmented curves are connected end to end at adjacent measured data points. Set mathematical constraints for the connection of each piecewise curve: at each internal node, two adjacent curve segments must have equal function values, first derivative values, and second derivative values. Based on all the data points and constraints mentioned above, by solving the system of linear equations, all coefficients of the mathematical expression for each interval segment are calculated, generating a reproductive kinetics curve and a growth curve that spans all measured time points.
[0014] Compared with the prior art, the advantages of the present invention are: This method achieves standardized and automated operation throughout the entire process, from sample collection and preparation to microscopic image acquisition and data analysis. By standardizing sampling locations, imaging illumination, and exposure parameters, it avoids detection errors caused by the randomness of traditional manual operations and differences in equipment parameters. Utilizing multi-dimensional image preprocessing, adaptive thresholding combined with precise edge detection segmentation technology, and a dual recognition mechanism of prior feature filtering and machine learning classifiers, it can efficiently remove interference from impurities in the koji material, accurately distinguish Aspergillus oryzae spores from interfering substances, and significantly improve the accuracy and stability of spore identification. Simultaneously, this method can acquire high-throughput multi-dimensional data such as spore concentration and morphological distribution. Through time-series data fitting, it generates high-precision reproductive kinetics and growth curves, accurately dividing spore reproduction stages and quantifying reproduction rates. This overcomes the shortcomings of traditional methods, such as low efficiency, strong subjectivity, and inability to finely characterize the dynamic reproduction patterns of spores, providing precise, objective, and continuous data support for the regulation of microbial growth during the koji-making process. Attached Figure Description
[0015] Figure 1 This is a schematic diagram of the Aspergillus oryzae spore reproduction analysis method based on microscopic image recognition proposed in this invention. Figure 2 This is a schematic diagram of the candidate spore connectivity regions selected according to the present invention. Figure 3 This is a schematic diagram illustrating the identification of Aspergillus oryzae spores and the generation of a contour mask proposed in this invention. Detailed Implementation
[0016] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.
[0017] See Figure 1 As shown, the method for analyzing the reproduction of Aspergillus oryzae spores during the koji-making process based on microscopic image recognition includes: At key stages of koji making, samples are collected from different locations of the koji material, and after low-temperature transfer and aseptic grinding, a uniform spore suspension is prepared. The suspension is allowed to stand and the supernatant is taken and quantitatively added to a standard counting plate to prepare a standardized sample with uniform thickness. The standardized sample is placed on the microscope stage, and multiple field-of-view positions are planned according to the preset grid. After focusing at each position, high-definition digital microscopic images are acquired under constant illumination and exposure parameters to form an image sequence covering the sample area. The image sequence was preprocessed with white balance correction, noise filtering and illumination homogenization. Contrast was enhanced by converting the color space. The image was binarized and segmented using an adaptive thresholding method combined with edge detection. Candidate spore connected regions were preliminarily screened through morphological operations. Based on the selected spore connectivity regions, the morphological, color, and texture features of each region are calculated. Preliminary filtering is performed based on the prior feature range of Aspergillus oryzae spores to remove obvious impurities. A pre-trained machine learning classifier is used to finely discriminate the remaining targets, identify Aspergillus oryzae spores, and generate contour masks. Based on the identification results, the spore concentration per unit volume is statistically analyzed, and the morphological distribution of the population spores is analyzed. The spore parameters at different time points are arranged in a time series to generate reproductive kinetic curves and growth curves, identify the reproductive stages, and calculate the reproductive rate of specific stages.
[0018] Specifically, the prepared standardized microscopic specimen is placed on the microscope stage, and the area to be scanned is defined. This area typically corresponds to a complete counting grid on a hemocytometer. Non-overlapping field-of-view locations are planned within this area according to a pre-defined grid matrix. The coordinates of each field-of-view location are... , ) from the starting point coordinates ( , The values are determined by the fixed step distance (ΔX, ΔY), and the calculation formula is as follows:
[0019]
[0020] The step distance (ΔX, ΔY) is slightly smaller than the actual field of view of the camera sensor under the corresponding objective lens to ensure a small amount of overlap between adjacent fields of view.
[0021] A sharpness evaluation function is adopted. To calculate the focus level of each preview image, a commonly used function is the sum of squares of the Laplacian operator based on the image gradient:
[0022] in, In the image The value calculated by applying the Laplacian operator at the mid-pixel (x, y). This function effectively responds to edges and details in an image, and its value reaches its maximum when the image is sharpest.
[0023] The focusing algorithm compares the corresponding positions at different locations. Find the Z-axis position that maximizes the evaluation function and move the stage accordingly. This position is the optimal focal plane for the field of view.
[0024] After autofocusing, the camera captures a high-resolution RGB color digital image of the current field of view under optimal focal plane, constant lighting, and uniform exposure parameters. This image is saved in a lossless or low-loss compression format to preserve the original information to the greatest extent possible.
[0025] Repeat the above steps until all planned field-of-view locations have been traversed. Finally, a sequence of high-resolution microscopic images is generated in the order of acquisition. While acquiring each image, a structured metadata file is automatically generated and associated, including the acquisition timestamp, field-of-view spatial coordinates, optical parameters, camera parameters, light source parameters, and image identifier.
[0026] See Figure 2 As shown, for each original RGB image in the image sequence Standardized preprocessing is performed to correct color deviations caused by fluctuations in light source color temperature or differences in camera response. An automatic white balance algorithm based on the gray-world assumption is used, assuming that the average reflection of the entire image is colorless. The image is then calculated separately. Average intensity values in the red (R), green (G), and blue (B) channels Using the green channel as a reference, calculate the gain coefficient:
[0027] The corrected image Iwb values for each channel are:
[0028] Image after white balance correction using a two-dimensional discrete Gaussian low-pass filter Smoothing is performed. For each pixel (x, y) in the image, its filtered grayscale value is... It is obtained by weighted averaging of its neighboring pixels, with the weights determined by a Gaussian function:
[0029] Where (u,v) are the coordinates relative to the center point, and σ is the standard deviation, which controls the degree of smoothness.
[0030] Background estimation and subtraction are used to correct the filtered image. Background estimation using morphological opening operation The opening operation is performed by first eroding and then dilating, using a structuring element much larger than the spore size to remove all foreground objects such as spores, retaining only the slowly changing background. (Corrected image) Obtained through the following formula:
[0031] Where C is a constant offset used to adjust the overall brightness of the image to the range of 0~255.
[0032] The homogenized RGB image The process is then converted to a color space more suitable for color analysis and segmentation, such as the HSV space. In the HSV space, the combined information from the saturation channel (S) and the lightness channel (V) of the spores effectively enhances their contrast with transparent impurities or dark residues. Subsequent segmentation is primarily based on the fusion of the S and V channels or processing using a single channel.
[0033] For a selected single-channel grayscale image (such as a fused SVSV channel or a single VV channel), denoted as IgrayIgray, an adaptive threshold segmentation algorithm combined with edge detection is used to binarize it.
[0034] grayscale image The area is divided into a series of overlapping local blocks. For a local window centered at (x, y), its initial local threshold is calculated. An effective method is to calculate a weighted average of the pixel grayscale values within the window, with pixels closer to the center having a higher weight. A simplified implementation uses a Gaussian weighted average.
[0035] in, It is a two-dimensional Gaussian weighting function. It iterates through all pixels in the entire image, generating a sum of all pixels and the sum of all pixels. Initial threshold maps of the same size.
[0036] For the same grayscale image An edge detection operator is applied to obtain an edge intensity map. The initial threshold may not be the optimal value for separating spores from the background at edge pixels and their neighborhoods. The threshold for a given point is adjusted based on the grayscale characteristics on both sides of the edge. For example, the grayscale values of pixels on both sides of the grayscale gradient direction within the edge point's neighborhood can be sampled, and their average value used as a more accurate local threshold. The local threshold is then used to replace the threshold for the corresponding edge region in the initial threshold map, forming an optimized, refined threshold map.
[0037] For grayscale images For each pixel in the image, its grayscale value is compared with the dynamic threshold at the corresponding position in the fine-grained thresholding image. Since spores are usually brighter than the background, if the grayscale value is greater than the dynamic threshold at the corresponding position in the fine-grained thresholding image, it is determined to be white; otherwise, it is determined to be black. This generates a preliminary binary image, in which spores appear as white connected regions. Opening and closing operations are then used to post-process the preliminary binary image, and candidate connected regions are extracted.
[0038] See Figure 3 As shown, for each candidate spore connected region marked in the morphologically processed binary image Where k = 1, 2, ..., K, and K is the total number of candidate regions, from the original color image The corresponding spatial location is used to extract its shape, color, and texture features, and a comprehensive feature vector is constructed.
[0039] The morphological features include region area, perimeter, equivalent diameter, aspect ratio, roundness, and convexity; the color features include the average intensity and intensity standard deviation of the connected regions in the original RGB or converted color space; and the texture features are obtained by calculating the local binary pattern. A rule-based filter based on statistical prior knowledge was established. Before deployment, a large number of known pure Aspergillus oryzae spore samples were collected using the same microscope, objective lens, and camera parameters, and spore regions were manually labeled and confirmed by experts. All the above features were calculated for the standard spore regions, and the distribution range of their key features was statistically analyzed.
[0040] For each candidate region And its feature vectors, applying concatenated judgment conditions, perform size filtering, shape filtering, and color filtering. If any condition is not met, then... It was identified as an obvious impurity and removed.
[0041] For the candidate target set that passes the fast filtering Its corresponding feature vector set The input is fed into a pre-trained machine learning classifier for final judgment, to distinguish spores with similar shapes and colors from impurities.
[0042] A large number of image patches were extracted from a historical image database and labeled as positive and negative samples by experts. The same d-dimensional feature vector was extracted for each sample to form a labeled training dataset. ,in These represent negative and positive samples, respectively.
[0043] Support Vector Machine (SVM) is used as the classifier. The goal of SVM is to find an optimal hyperplane. To maximize the margin between positive and negative samples. Its decision function is:
[0044] in, It is a nonlinear mapping implemented through kernel functions, mapping features to a high-dimensional space to achieve linear separability. The optimal solution is obtained by solving an optimization problem. And b, the model can be based on The symbols and values are classified.
[0045] For the candidate target to be tested eigenvectors This data is then fed into a trained SVM model. The model outputs not only class labels but also a discrimination score related to the distance to the decision hyperplane. A high-confidence threshold is set; only when the discrimination score exceeds this threshold is the candidate target finally included in the test. It was identified as a confirmed Aspergillus oryzae spore.
[0046] Iterate through all candidate regions. For each region ultimately identified as a spore, mark the corresponding connected region as valid at its original binary image location. Merge the binary masks of all valid spore regions to generate a final Aspergillus oryzae spore target mask image. Simultaneously, output a list of feature vectors and their spatial locations for each confirmed spore.
[0047] After identifying and locating Aspergillus oryzae spores at each koji-making time point, the spore concentration per unit volume and population morphology parameters were calculated. For samples collected at each time point, a set of spore target mask images was confirmed after image analysis. For each mask image, the number of connected regions marked as spores was counted, i.e., the spore count for that field of view. The spore count per field of view was then converted into the spore concentration per unit volume of sample solution.
[0048] The average and standard deviation of the concentrations in all valid fields of view at the same time point were calculated as the spore density of Aspergillus oryzae and its measurement dispersion at that koji-making time point. At the same time, the morphological characteristics of all confirmed spores at that time point were analyzed as an indicator of spore growth status.
[0049] Different music production time points Spore density obtained and average diameter Organized into time series datasets and To obtain a continuous, smooth dynamic curve that reflects the inherent trend of the data, a cubic spline interpolation method is used, interpolating between every two adjacent time nodes. Construct a cubic polynomial within the interval. To fit the data change trend within this interval. For the spore density curve, we have:
[0050] For the average spore diameter curve, there are curves with the same form. The coefficient is .
[0051] To ensure the smoothness of the entire curve, it is required at each internal node... At a given point, two adjacent polynomial segments satisfy continuity in function value, first derivative, and second derivative. Taking the spore density curve as an example, the constraint condition is: Function values are continuous: The first derivative is continuous: The second derivative is continuous: In addition, boundary conditions need to be specified at the first and last nodes.
[0052] The above constraints, along with the condition that each interval segment must pass through the endpoint data, together constitute a condition concerning all unknown coefficients. A system of linear equations. By solving this system of linear equations, the coefficients of the cubic polynomials over all intervals can be uniquely determined. Ultimately, the entire reproductive kinetic curve is defined as a piecewise function. : Similarly, a growth curve of average spore size can be generated. .
[0053] Based on the generated continuous and smooth reproductive kinetic curves, the derivatives in their logarithmic form or direct analysis are used to analyze... The changing trends are used to objectively divide the reproductive stages.
[0054] The germination / retardation phase is the stage in which C(t) remains at a low level and grows extremely slowly, typically corresponding to... ; The rapid reproductive / log phase is the stage where C(t) exhibits approximately exponential growth, during which lnC(t) is approximately linear. It reaches its maximum value and remains relatively stable. The stable period is the stage where C(t) reaches its peak and remains basically unchanged or fluctuates slightly, corresponding to... .
[0055] By searching The system can automatically determine the start and end times of each stage by setting the inflection point or a threshold value.
[0056] During the identified rapid reproduction period, the average reproduction rate was calculated. The rate is a key parameter characterizing the match between Aspergillus oryzae cell viability and koji-making process conditions.
[0057] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are possible or may be advantageous.
[0058] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.
[0059] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A method for analyzing the reproduction of Aspergillus oryzae spores during the koji-making process based on microscopic image recognition, characterized in that, include: At key stages of koji making, samples are collected from different locations of the koji material, and after low-temperature transfer and aseptic grinding, a uniform spore suspension is prepared. The suspension is allowed to stand and the supernatant is taken and quantitatively added to a standard counting plate to prepare a standardized sample with uniform thickness. The standardized sample is placed on the microscope stage, and multiple field-of-view positions are planned according to the preset grid. After focusing at each position, high-definition digital microscopic images are acquired under constant illumination and exposure parameters to form an image sequence covering the sample area. The image sequence was preprocessed with white balance correction, noise filtering and illumination homogenization. Contrast was enhanced by converting the color space. The image was binarized and segmented using an adaptive thresholding method combined with edge detection. Candidate spore connected regions were preliminarily screened through morphological operations. Based on the selected spore connectivity regions, the morphological, color, and texture features of each region are calculated. Preliminary filtering is performed based on the prior feature range of Aspergillus oryzae spores to remove obvious impurities. A pre-trained machine learning classifier is used to finely discriminate the remaining targets, identify Aspergillus oryzae spores, and generate contour masks. Based on the identification results, the spore concentration per unit volume is statistically analyzed, and the morphological distribution of the population spores is analyzed. The spore parameters at different time points are arranged in a time series to generate reproductive kinetic curves and growth curves, identify the reproductive stages, and calculate the reproductive rate of specific stages.
2. The method for analyzing the reproduction of Aspergillus oryzae spores during the koji-making process based on microscopic image recognition according to claim 1, characterized in that, The image sequence that forms the sample coverage area specifically includes: Place the prepared standardized sample on the microscope stage; Within the sample observation area, multiple non-overlapping field-of-view locations are planned according to a pre-defined grid matrix; At each preset field-of-view location, the microscope autofocus system executes a focusing algorithm based on image contrast to acquire the optimal focal plane image; Under constant light intensity and uniform exposure parameters, the camera sequentially acquires raw RGB color microscopic digital images of each field of view, forming a high-throughput image sequence covering the sample area; Record metadata for each image, including acquisition time, field of view coordinates, objective magnification, and light source parameters.
3. The method for analyzing the reproduction of Aspergillus oryzae spores during the koji-making process based on microscopic image recognition according to claim 1, characterized in that, The candidate spore connectivity regions selected specifically include: The acquired high-throughput raw microscopic image sequences were standardized and preprocessed, and all raw images were uniformly white-balanced to eliminate color temperature differences. A Gaussian filtering-based method is used for noise suppression, which smooths the image background while preserving spore edge details. To address potential uneven lighting, a background subtraction algorithm is applied to obtain an image with uniform brightness. The preprocessed RGB image is converted to a color space to enhance the contrast between the spores and the background and impurities; An adaptive threshold segmentation algorithm, combined with the assistance of an edge detection operator, is used to dynamically determine the optimal segmentation threshold for each image or local region of an image, thereby generating a binarized image. Morphological operations are performed on the binarized image. Opening operations are used to remove small noise points, and closing operations are used to fill the cavities inside the spores and smooth the spore outlines, thus initially obtaining the connected regions of candidate spores.
4. The method for analyzing the reproduction of Aspergillus oryzae spores during the koji-making process based on microscopic image recognition according to claim 3, characterized in that, The specific process for generating a binary image includes: The preprocessed grayscale image is divided into multiple overlapping or adjacent local regions. For each local region, the weighted average of the pixel grayscale values within the region is calculated as the initial local threshold for the center pixel position. An initial threshold map is generated by traversing the entire image. The edge detection operator is used to scan the image to identify edge pixels where the gray level changes abruptly at the boundary between the spore and the background. The thresholds at the corresponding positions in the initial threshold map are then fine-tuned and corrected to obtain an optimized fine threshold map. For each pixel in the original image, the dynamic segmentation threshold for that pixel location is read from the fine threshold map based on its coordinates. The gray value of the pixel is compared with the dynamic threshold to divide the foreground and background, generating a preliminary binarized image. Aspergillus oryzae spores appear as white connected regions, while the background is black.
5. The method for analyzing the reproduction of Aspergillus oryzae spores during the koji-making process based on microscopic image recognition according to claim 1, characterized in that, The process of identifying Aspergillus oryzae spores and generating a contour mask includes: For each candidate spore connected region obtained, its morphological, color, and texture features are calculated, and a feature vector is constructed to describe it; The morphological features include region area, perimeter, equivalent diameter, aspect ratio, roundness, and convexity; the color features include the average intensity and intensity standard deviation of the connected regions in the original RGB or converted color space; and the texture features are obtained by calculating the local binary pattern. A feature filter based on prior knowledge is set up to initially screen candidate targets according to the typical feature range of pure Aspergillus oryzae spores under the microscopic conditions used, and obvious non-spore targets with irregular size, shape and color deviation are eliminated. For targets that pass the initial screening, a classifier based on a pre-trained machine learning model is used for fine classification. The model is trained using labeled positive samples of Aspergillus oryzae spores and negative samples of non-spores. The final output is a target mask image that is confirmed to be Aspergillus oryzae spores.
6. The method for analyzing the reproduction of Aspergillus oryzae spores during the koji-making process based on microscopic image recognition according to claim 5, characterized in that, The aforementioned feature filter, based on prior knowledge, performs preliminary screening of candidate targets according to the typical characteristic range of pure Aspergillus oryzae spores under the applied microscopic conditions, eliminating obvious non-spore targets with irregular size, shape, or color deviation. Specifically, this includes: Under the same microscopic imaging conditions, images of known pure Aspergillus oryzae spores were pre-acquired and analyzed, the distribution range of the population on key features was statistically analyzed, and a prior knowledge base was constructed. Based on the constructed prior knowledge base, a series of logical judgment conditions are set to eliminate candidate targets whose equivalent diameter is smaller or larger than the normal range of spores, eliminate candidate targets whose roundness is lower than the threshold or whose length-to-short axis ratio is higher than the threshold, and exclude candidate targets that deviate from the typical range of spores in terms of key color features. The feature vector of each candidate target is compared with the multi-level filtering rules in turn, and the candidate targets that pass all logical judgment conditions are retained.
7. The method for analyzing the reproduction of Aspergillus oryzae spores during the koji-making process based on microscopic image recognition according to claim 5, characterized in that, The classifier of the pre-trained machine learning model specifically includes: Annotated image patches are extracted from historical microscopic images to construct a training dataset. Positive samples are image patches containing complete and typical Aspergillus oryzae spores, while negative samples contain interference objects with similar morphology or color to spores. For each training sample image patch, a multi-dimensional feature vector containing all morphological, color, and texture features is generated. The model is trained by inputting multi-dimensional feature vectors using a machine learning algorithm based on support vector machines. After training, the model performance is evaluated using a validation set. For candidate targets that pass the initial screening, extract the image patch of the region where it is located, and calculate the same multi-dimensional feature vector and input it into the pre-trained model to calculate the probability that the target belongs to the Aspergillus oryzae spore category, and determine whether the target is an Aspergillus oryzae spore; For all targets identified as spores, the corresponding connected regions in their binarized mask images are marked as valid spore regions, ultimately generating a target mask image containing only the identified spores.
8. The method for analyzing the reproduction of Aspergillus oryzae spores during the koji-making process based on microscopic image recognition according to claim 1, characterized in that, The generation of reproductive kinetic curves and growth curves, identification of reproductive stages, and calculation of the reproductive rate at specific stages specifically include: Based on the Aspergillus oryzae spore targets confirmed by fine screening, the number of spores identified in each microscopic image was calculated; By combining the microscope objective magnification, camera pixel size, and known volume depth of the counting chamber, the number of spores per field of view is converted into the spore concentration per unit volume of sample solution; The spore concentrations of all effective fields of view of the current sample are statistically analyzed, and their average value and standard deviation are calculated as the Aspergillus oryzae spore density at this koji-making time point. Analyze the distribution of morphological characteristics of all identified spores; The parameters obtained at different music production time points are arranged in a time series. The spline interpolation method was used to generate the reproductive kinetic curve of Aspergillus oryzae spore density over time and the growth curve of average spore size over time during the koji-making process. Based on the reproductive kinetics curve and growth curve, the spore germination period, rapid reproduction period and stationary period are identified, and the reproduction rate of specific stages is calculated.
9. The method for analyzing the reproduction of Aspergillus oryzae spores during the koji-making process based on microscopic image recognition according to claim 8, characterized in that, The specific methods for generating the reproductive kinetics curve of Aspergillus oryzae spore density over time and the growth curve of average spore size over time using spline interpolation include: The measured data obtained through microscopic image analysis at different time points in the koji-making process were collected to form a time-spore density dataset and a time-spore average diameter dataset. The interval between adjacent measured time points is defined as a segment, and a smooth mathematical curve is constructed independently for each segment. The segmented curves are connected end to end at adjacent measured data points. Set mathematical constraints for the connection of each piecewise curve: at each internal node, two adjacent curve segments must have equal function values, first derivative values, and second derivative values. Based on all the data points and constraints mentioned above, by solving the system of linear equations, all coefficients of the mathematical expression for each interval segment are calculated, generating a reproductive kinetics curve and a growth curve that spans all measured time points.