Accurate identification method and program product of toxin-producing fungi based on intelligent processing of image contour temporal sequence

Abstract

The present invention relates to the technical field of mycotoxin detection in agricultural products, and in particular to a method and program product for accurately identifying toxin-producing fungi based on intelligent temporal processing of image contours. The method comprises the following steps: 1) preprocessing micrographs; 2) extracting fungal color features and converting the micrographs into grayscale images; 3) extracting fungal edge lines; 4) converting the fungal image edge lines into time-series data; 5) performing shapelet-based centralized learning on the fungal image edge line time-series data to obtain a shapelet edge line feature set; 6) calculating a similarity matrix between the fungal micrograph samples to be identified and the shapelet edge line feature set obtained in step 5) through edge line extraction and temporal processing; and 7) using this similarity matrix, combined with secondary verification using a fungal feature knowledge base, to achieve intelligent identification and classification of fungal micrographs. This method achieves rapid, accurate, and efficient results.

Description

Technical Field

[0001] The present invention relates to the technical field of mycotoxin detection in agricultural products, and in particular to a method and program product for accurately identifying toxin-producing fungi based on intelligent processing of image contour temporal sequences. Background Art

[0002] According to the Food and Agriculture Organization of the United Nations (FAO), 25% of agricultural products worldwide are contaminated with mycotoxins each year. In my country, the impact of mycotoxin contamination on agricultural products is even more severe due to individual farmers' planting and storage methods, the high temperature and humidity in the Yangtze River Basin and South China, and consumer habits.

[0003] Agricultural products produce and accumulate various mycotoxins during the growth, harvesting, processing, storage, transportation, and sales processes. In particular, as storage and transportation time increases, molds are more likely to infect and reproduce, and once the temperature and humidity conditions are suitable, mycotoxins will accumulate in large quantities. Identifying key toxin-producing fungi is the basis and key to identifying, monitoring, and controlling the risks of mycotoxins in agricultural products. Toxin-producing fungi can be identified through laboratory testing, but due to the wide variety of fungi and their different morphologies, observing and measuring the morphology, size, and number of fungi under a microscope can easily cause visual fatigue and inaccurate measurements. In addition, samples are often mixed with other microorganisms, which greatly interferes with the identification of fungi using traditional microscopes, increasing the difficulty of fungal identification.

[0004] A Chinese invention patent application (CN110793946A, publication date 2020-02-14) discloses a fungal sample microscopic imaging and intelligent identification system and method. The system includes: a microscopic imaging device, a three-dimensional mobile control platform and a terminal; the microscopic imaging device is used to provide an illumination light source for microscopic imaging and detect the fluorescent signal emitted by the sample to be tested to obtain a fungal microscopic image of the sample to be tested; the three-dimensional mobile control platform is used to carry the sample to be tested to move to achieve the functions of automatic focusing and scanning imaging; the terminal is used to control the microscopic imaging device and the three-dimensional mobile control platform to work together, obtain fungal microscopic images from the microscopic imaging device, analyze the fungal microscopic images, and depict the outline of the hyphae structure in the fungal microscopic images.

[0005] Furthermore, a Chinese invention patent application (CN110796661A, published on February 14, 2020) discloses a method and system for segmenting and detecting fungal microscopic images based on a convolutional neural network. The method comprises: dividing a number of fungal microscopic images into positive samples and negative samples, while simultaneously labeling the hyphae in the positive samples to obtain labeled positive samples; slicing and performing sample enhancement operations on the unlabeled negative samples and the labeled positive samples to generate a training dataset for deep learning; constructing a deep convolutional neural network model, reading the training dataset to generate a segmentation model for segmentation detection, and using the segmentation model to identify the pathogenic and non-pathogenic phases in the fungal microscopic image to be detected. The global fungal microscopic image is represented by a thermal map, and the outline of the pathogenic phase structure in the fungal microscopic image to be detected is depicted. This patent not only classifies fungal microscopic images, but also enables the segmentation and precise location of the hyphal structure of the pathogenic phase from the image.

[0006] A Chinese invention patent application (CN110807754A, published on February 18, 2020) discloses a method and system for segmenting and detecting fungal microscopic images based on deep semantic segmentation. The method comprises: collecting N fungal microscopic images, removing images with entirely black backgrounds from the N fungal microscopic images to obtain a remaining image, and marking positive areas in the remaining image to obtain a marked image; slicing the remaining image and the marked image based on information from the marked image to generate a required training dataset; constructing a deep convolutional neural network model using a residual network and a dilated convolution module as the main framework of the network, reading the training dataset to train and generate a target model for segmentation detection, and using the target model to identify the fungal microscopic image to be detected to obtain a segmentation result of the pathogenic phase of the fungus in the fungal microscopic image to be detected. This patent enables the hyphal structure of the pathogenic phase to be segmented and accurately located from the image.

[0007] Time series analysis and prediction is a qualitative analysis method that uses mathematical methods to establish a prediction model based on time series variable analysis, extending time trends outward to predict market development and change trends and determine variable prediction values. It is also called time series analysis, historical extension method, and extrapolation method. Sequence analysis is very useful in applications such as stock market analysis, weather forecasting, and product recommendations. The two public documents mentioned above each used annotated images for slicing, generated the required training data sets, and performed manual judgment. Currently, there is no technology that uses the time series analysis of this application to accurately measure and automatically identify fungi. Summary of the Invention

[0008] To address the problems of inaccuracy, interference, and visual fatigue that plague traditional observation and identification methods due to the large variety and diverse morphologies of fungi, the present invention aims to provide a method for accurately identifying toxin-producing fungi based on intelligent processing of image contour temporal sequences. This method utilizes computer technology and algorithmic models to analyze and understand microscopic fungal images. Through specific technical steps, this method enables precise measurement and automatic identification of toxin-producing fungi, achieving rapid, accurate, and efficient results.

[0009] In order to achieve the above-mentioned purpose, the present invention adopts the following technical solutions:

[0010] A method for accurately identifying toxin-producing fungi based on intelligent processing of image contour temporal sequence, characterized in that the method comprises the following steps:

[0011] 1) Micrograph preprocessing;

[0012] 2) Extract fungal color features and convert micrographs into grayscale images;

[0013] 3) Extract the fungal edge line;

[0014] 4) Convert the edge lines of fungus photos into time series data;

[0015] 5) Perform shapelet-based learning on the time series data of the edge lines of fungus photos to obtain the shapelet edge line feature set;

[0016] 6) For the fungus micrograph sample to be identified, calculate the similarity matrix with the Shapelet edge feature set in step 5) by extracting and temporizing the edge lines;

[0017] 7) Based on the above similarity matrix and combined with secondary proofreading of the fungal feature knowledge base, intelligent recognition and classification of fungal micrographs are obtained.

[0018] Preferably, the step 2) uses the cvtColor method to convert the image into a grayscale image, and the processing function is:

[0019] ;

[0020] Where (x, y) is the coordinate point in the image, and the function is the value of the corresponding coordinate point.

[0021] Preferably, in step 3), the GaussianBlur method is used to smooth the boundary to increase the width of the boundary, and the Canny operator is used to extract the fungus edge line feature.

[0022] Preferably, in step 4), N pixels are uniformly sampled clockwise on the edge line of the fungus microscopic image, the center point of the minimum circumscribed rectangle of the fungus image is selected as a reference point, and the distances from the N sampled pixels to the center point are calculated to obtain time series data.

[0023] Preferably, in step 6), the fungus microscopic image to be identified is subjected to preprocessing including screenshot, grayscale conversion, denoising, edge line extraction, and edge line sampling pixel points in step 4) to obtain time series data of the fungus image to be identified; and then the data is calculated with the Shapelet edge line feature set in step 5) to obtain a similarity matrix value.

[0024] Preferably, the specific steps of step 6) include the following process:

[0025] 6.1) The fungus microscopic image to be identified consists of N samples, whose time series data is T = {T0, T1, …, TN-1}. The Shapelet edge feature set in step 5) is a set of M Shapelet time series S = {S0, S1, …, SM-1}.

[0026] 6.2) Calculate the Z-score of a single time series Ti and Sj. The Z-score standardization conversion formula is as follows: ,in, , , the similarity measure uses Euclidean distance and is normalized by the length of Shapelet; a similarity matrix Di,j of dimension N*M is obtained, where Di,j represents the similarity distance between the i-th sample and the j-th Shapelet time series in the Shapelet edge line feature set.

[0027] As a further preferred embodiment, the similarity distance Di,j between the i-th sample and the j-th Shapelet time series in the Shapelet edge line feature set is calculated as follows:

[0028] (1) Assume that the time series data of a sample is Τi = {t0, t1, …, tm}, and any time series in the Shapelet edge feature set is Sj = {s0, s1, …, sn}, where m and n are the lengths of the two sequences respectively;

[0029] (2) Calculate the Zscore of the two sequences respectively and obtain the normalized sequence data ZTi and ZSj;

[0030] (3) Enumerate all starting points in the time series Τi , for each substring in the time sequence Τi Calculate the Euclidean distance between it and the current Shapelet Sj , select all the starting points and calculate all the distances ; Let the similarity distance between the time series Ti and the j-th Shapelet time series Sj be defined as .

[0031] Furthermore, the present invention also discloses a computer device, comprising a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the method.

[0032] Furthermore, the present invention also discloses a computer-readable storage medium having a computer program or instruction stored thereon, which implements the method when the computer program or instruction is executed by a processor.

[0033] Furthermore, the present invention also discloses a computer program product, comprising a computer program or instructions, which implement the method when executed by a processor.

[0034] By adopting the aforementioned technical solution, the present invention utilizes computer technology and algorithmic models to analyze and understand microscopic fungal images. Through specific technical steps, it achieves precise measurement and automatic identification of toxin-producing fungi, achieving rapid, accurate, and efficient results. This system can be applied in related fields such as the identification, monitoring, and control of mycotoxin risks in agricultural products. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 It is a flow chart of the method of the present invention.

[0036] Figure 2 This is a microscopic photograph of fungi according to one embodiment of the present invention.

[0037] Figure 3 This is a grayscale rendering of a captured image according to a specific embodiment of the present invention.

[0038] Figure 4 This is an edge line map extracted according to a specific embodiment of the present invention.

[0039] Figure 5 This is a schematic diagram of the intelligent recognition and classification after inputting the contour of the present invention. DETAILED DESCRIPTION

[0040] The following is a detailed description of the specific embodiments of the present invention with reference to the accompanying drawings.

[0041] like Figure 1 The following is a method for accurately identifying toxin-producing fungi based on intelligent processing of image contour temporal data:

[0042] (Step 1) Microscopic image preprocessing

[0043] Select a relatively clear fungus from the microphotograph and capture the image (e.g. Figure 2 The rectangular area (shown in the image) must completely encompass the fungus and its edges. Image capture is done to filter out invalid image portions, such as those containing noise, blurred images, overlapping or incomplete images, thereby improving the image's signal-to-noise ratio.

[0044] (Step 2) Extract fungus-related color features and convert the image into a grayscale image using the cvtColor method.

[0045] Fungal micrographs are usually large files, at least several MB. Directly extracting feature parameters from them will be slow, so we first perform grayscale processing on them and convert them into a grayscale image matrix. The function for grayscale processing is:

[0046] ;

[0047] Where (x, y) is the coordinate point in the image, and the function is the value of the corresponding coordinate point.

[0048] The effect of the captured image after being grayscaled is shown in Figure 3.

[0049] (Step 3) Image denoising and fungus edge feature extraction

[0050] The GaussianBlur method (Gaussian filtering) is used to smooth the boundary and increase its width. The boundary feature is the large difference in the value on the left and right sides of a certain boundary. We use the Canny operator to extract the fungus edge line feature.

[0051] The input of a Gaussian filter is a grayscale image matrix, and the output is a filtered grayscale image matrix. Gaussian filtering is a process of weighted averaging the entire image. The value of each pixel is the weighted average of its own value and the values ​​of other pixels in its neighborhood.

[0052] Common edge extraction methods include the Canny operator extraction method, the Sobel operator extraction method, the Roberts operator extraction method, etc. Among them, the Canny() function extraction method takes as input the binary image matrix to be edge detected and the edge extraction method, and outputs the binary image matrix after edge extraction.

[0053] The extracted edge lines are as follows Figure 4 shown.

[0054] (Step 4) Convert the edge lines of the fungus image into time series data. Sample N pixels uniformly along the edge lines of the fungus micrograph in a clockwise direction. Select the center point of the smallest circumscribed rectangle of the fungus image as the reference point. Calculate the distances from each of the N sampled pixels to this center point to generate the time series data.

[0055] (Step 5) Perform shapelet centralized learning on the fungal edge line time series data to obtain the shapelet edge line feature set.

[0056] (Step 6) For the fungus microscopic image to be identified, preprocessing is performed, including screenshots, grayscale conversion, denoising, and edge extraction. This is combined with the edge line sampling pixel points from Step 4 to generate time series data for the fungus image to be identified. This data is then combined with the Shapelet edge line feature set from Step 5 to generate a similarity matrix.

[0057] The fungal microscopic image to be identified consists of N samples, whose time series data is T={T0,T1,…,TN-1}. The Shapelet edge feature set in step 5 is a set of M Shapelet time series S={S0,S1,…,SM-1}. The Z-score of each time series Ti and Sj is calculated. The Z-score standardization conversion formula is as follows: ,in, , , the similarity measure uses Euclidean distance and is normalized by the length of Shapelet; a similarity matrix D of dimension N*M is obtained, where Di,j represents the similarity distance between the i-th sample and the j-th Shapelet time series in the Shapelet edge line feature set.

[0058] The calculation process of the similarity distance Di,j between the i-th sample and the j-th Shapelet time series in the Shapelet edge line feature set is as follows: (1) Assume that the time series data of a certain sample is Τi= {t0, t1, …, tm}, and any time series in the Shapelet edge line feature set is Sj= {s0, s1, …, sn}, where m and n are the lengths of the two sequences respectively; (2) Calculate the Zscore of the two sequences respectively and obtain the normalized sequence data ZTi and ZSj; (3) Enumerate all the starting points in the time series Τi , for each substring in the time sequence Τi Calculate the Euclidean distance between it and the current Shapelet Sj , select all the starting points and calculate all the distances ; Let the similarity distance between the time series Ti and the j-th Shapelet time series Sj be defined as .

[0059] (Step 7) The similarity matrix in step 6 is combined with the secondary proofreading of the fungal feature knowledge base, including feature points such as fungal length / width, perimeter, circularity, and rectangularity, to obtain intelligent classification of fungal microscopic images.

[0060] The above is a description of the embodiments of the present invention. The above description of the disclosed embodiments will enable professionals in the field to implement or use the present invention. Various modifications to these embodiments will be apparent to professionals in the field. The general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for accurately identifying toxin-producing fungi based on intelligent processing of image contour temporal sequence, characterized in that: The method comprises the following steps: 1) Micrograph preprocessing; 2) Extract fungal color features and convert micrographs into grayscale images; 3) Extract the fungal edge line; 4) Convert the edge lines of fungus photos into time series data; 5) Perform shapelet-based learning on the time series data of the edge lines of fungus photos to obtain the shapelet edge line feature set; 6) For the fungus micrograph sample to be identified, calculate the similarity matrix with the Shapelet edge feature set in step 5) by extracting and temporizing the edge lines; 7) Based on the similarity matrix and secondary proofreading of the fungal feature knowledge base, intelligent recognition and classification of fungal micrographs are achieved; In step 4), N pixels are evenly sampled clockwise on the edge line of the fungus microscopic image, the center point of the minimum circumscribed rectangle of the fungus image is selected as a reference point, and the distances from the N sampled pixels to the center point are calculated to obtain time series data.

2. The identification method according to claim 1, characterized in that Step 2) Use the cvtColor method to convert the image into a grayscale image. The processing function is: ； Where (x, y) is the coordinate point in the image, and the function is the value of the corresponding coordinate point.

3. The identification method according to claim 1, characterized in that Step 3) Use the GaussianBlur method to smooth the boundary and increase the width of the boundary, and use the Canny operator to extract the fungus edge line features.

4. The identification method according to claim 1, wherein: In step 6), the fungus microscopic image to be identified is subjected to preprocessing including screenshot, grayscale conversion, denoising, edge line extraction, and edge line sampling pixel points in step 4) to obtain time series data of the fungus image to be identified; this data is then calculated with the Shapelet edge line feature set in step 5) to obtain a similarity matrix value.

5. The identification method according to claim 4, characterized in that: The specific steps of step 6) include the following process: 6.1) The microscopic image of the fungus to be identified consists of N samples, and the time series data of the N samples is Τ={T0,T1,…,T N-1 }, the Shapelet edge feature set in step 5) is a set of M Shapelet time series S={S0,S1,…,S M-1 }; 6.2) Calculate a single time series T i and S j The Z-score, Z-score standardization conversion formula is as follows ， in, ， ， The similarity metric uses Euclidean distance and is normalized by the length of the Shapelet; the similarity matrix D with a dimension of N*M is obtained. i,j , where D i,j Represents the similarity distance between the i-th sample and the j-th Shapelet time series in the Shapelet edge line feature set.

6. The identification method according to claim 5, characterized in that The similarity distance D between the i-th sample and the j-th Shapelet time series in the Shapelet edge line feature set i,j The calculation process is: (1) Assume that the time series data of a sample is T i = {t0,t1,…,t m }, any time sequence in the Shapelet edge line feature set is S j = {s0,s1,…,s n }, where m and n are the lengths of the two sequences respectively; (2) Calculate the Z of the two sequences separately score , and obtain the normalized sequence data Z Ti and Z Sj ; (3) Enumeration time series Τ i All starting points in ， For the timing Τ i Each substring in Calculate it and the current Shapelet S j Euclidean distance ， Select all starting points and calculate all distances ； Let the timing sequence T i and the j-th Shapelet sequence S j The similarity distance between them is defined as 。 7. A computer device comprising a memory, a processor, and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the method according to any one of claims 1 to 6.

8. A computer-readable storage medium having a computer program or instruction stored thereon, characterized in that: When the computer program or instruction is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

9. A computer program product comprising a computer program or instructions, characterized in that When the computer program or instruction is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images