Microorganism efficient detection method and device based on image recognition
By employing an image recognition-based microbial detection method, and utilizing automation technology and deep learning models, the problems of low efficiency and poor accuracy in traditional microbial detection have been solved. This method enables efficient differentiation between damaged microorganisms and contaminated areas, thereby improving detection efficiency and accuracy.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-13
- Publication Date
- 2026-03-10
AI Technical Summary
Traditional microbial detection methods are inefficient, difficult to improve accuracy, and easily affected by subjective factors.
An efficient microbial detection method based on image recognition is adopted. By utilizing automated image recognition technology and deep learning models, it can distinguish between damaged microorganisms and contaminated areas through region of interest identification, complete microbial region extraction, damage morphology fitting and cluster analysis, thereby improving detection efficiency and accuracy.
It significantly improves the efficiency and accuracy of microbial detection, reduces data processing volume and false positives for contamination, and enables efficient differentiation between damaged microorganisms and contaminated areas.
Smart Images
Figure CN121640458A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the field of data processing, in particular to a microorganism efficient detection method and device based on image recognition. BACKGROUND
[0002] With the increasing application of microorganism research in the fields of environmental monitoring, food safety, medical health, etc., microorganism detection technology becomes increasingly important. Traditional microorganism detection methods usually rely on microscope observation and manual counting, which not only consumes time and effort, but also is easily affected by subjective factors, making it difficult to improve the accuracy of detection results.
[0003] Therefore, in the prior art, the microorganism detection method has the technical problems of low detection efficiency and difficult to improve detection accuracy. SUMMARY
[0004] The present application provides a microorganism efficient detection method and device based on image recognition, which solves the technical problems of low detection efficiency and difficult to improve detection accuracy in the prior art. Through the automatic image recognition technology and deep learning model, the damaged microorganisms and the stain areas are efficiently distinguished, the data processing amount of subsequent microorganism detection is significantly reduced, the stain misrecognition situation is reduced, and the efficiency and accuracy of microorganism detection are improved.
[0005] The present application provides a microorganism efficient detection method based on image recognition, which comprises: collecting a microorganism detection image sequence for microorganism detection, performing region of interest recognition on a plurality of microorganism detection images in the microorganism detection image sequence, and obtaining a plurality of region of interest sets. Perform complete microorganism recognition on the plurality of regions of interest to obtain a plurality of complete microorganism region sets, delete the plurality of complete microorganism region sets in the plurality of region of interest sets to obtain a plurality of key region sets. Based on a plurality of damage morphologies of microorganism damage, fitting the pixel points in the plurality of key region sets to obtain a plurality of fitting result sets, and screening to obtain a plurality of assumed damaged microorganism region sets and a plurality of assumed stain region sets. Cluster the plurality of assumed damaged microorganism region sets to obtain a plurality of assumed damaged microorganism region image sequences of a plurality of assumed damaged microorganisms. Perform microorganism dynamic recognition on the plurality of assumed damaged microorganism region image sequences to obtain a plurality of damaged microorganism region sets and a plurality of actual stain region sets, add the plurality of assumed stain region sets to obtain a plurality of stain region sets, and perform microorganism detection on the plurality of damaged microorganism region sets and the plurality of complete microorganism region sets.
[0006] In an implementation, a sequence of microbe detection images for microbe detection is collected, a plurality of region of interests in the sequence of microbe detection images are identified, and a plurality of region of interest sets are obtained. The collection of sample microbe detection images is collected based on historical data of microbe detection, a region of interest in each sample microbe detection image is identified and labeled, and a collection of sample region of interest division results is obtained, where the region of interest includes an image region of a microbe and a stain. The region of interest identifier is trained using the collection of sample microbe detection images and the collection of sample region of interest division results. The region of interest identifier is used to identify the plurality of regions of interest in the sequence of microbe detection images, and the plurality of region of interest sets are obtained.
[0007] In an implementation, the plurality of complete microbe region sets are deleted from the plurality of region of interest sets to obtain a plurality of key region sets, including: collecting a sample region of interest set and labeling whether a region of interest includes a complete microbe image to obtain a sample complete classification information set, where the sample complete classification information is 1 or 0. The sample region of interest set and the sample complete classification information set are used to train a complete microbe region identifier to identify the plurality of region of interest sets to obtain a plurality of complete classification information sets. The region of interest corresponding to the complete classification information of 1 in the plurality of complete classification information sets is extracted as a plurality of complete microbe region sets. The plurality of complete microbe region sets are deleted from the plurality of region of interest sets to obtain a plurality of key region sets.
[0008] In an implementation, based on a plurality of damage morphologies of microbe damage, pixel points in the plurality of key region sets are fitted to obtain a plurality of fitting result sets, and a plurality of assumed damaged microbe region sets and a plurality of assumed stain region sets are obtained, including: a plurality of fitting boxes are obtained based on a plurality of damage morphologies of microbe damage. The pixel points in the plurality of key region sets are iteratively fitted using the plurality of fitting boxes to obtain a plurality of fitting result sets, where each fitting result includes a microbe fitting degree. It is determined whether the microbe fitting degree in each fitting result is greater than a fitting degree threshold. If yes, the corresponding key region is an assumed damaged microbe region, and if no, the corresponding key region is an assumed stain region, and the plurality of assumed damaged microbe region sets and the plurality of assumed stain region sets are obtained.
[0009] In an implementation, the pixel points in the plurality of key region sets are iteratively fitted using the plurality of fitting boxes to obtain a plurality of fitting result sets, including: randomly selecting a first fitting box in the plurality of fitting boxes, and performing random box fitting in a first key region in the plurality of key region sets to obtain a proportion of pixel points in the first key region falling into the first fitting box as a first fitting degree. Continue to fit the first key region using other fitting boxes to obtain a plurality of fitting degrees, output the maximum fitting degree as a first microorganism fitting degree, and obtain a first fitting result. Continue to iteratively fit the plurality of key region sets using the plurality of fitting boxes to obtain a plurality of microorganism fitting results as the plurality of fitting result sets.
[0010] In an implementation, the plurality of assumed damaged microorganism region sets are clustered to obtain a plurality of assumed damaged microorganism region image sequences of a plurality of assumed damaged microorganisms, including: randomly combining the assumed damaged microorganism regions in the plurality of assumed damaged microorganism region sets, performing image similarity analysis to obtain a plurality of assumed damaged microorganism similarity sets. Obtain a preset moving region, the preset moving region being a region range of the same microorganism moving in the microorganism detection image sequence. Screen the assumed damaged microorganism regions in the same preset moving region in the plurality of microorganism detection images, and the damaged microorganism similarity is greater than the similarity threshold value, complete the clustering of the damaged microorganism region, and sort according to the time sequence to obtain a plurality of assumed damaged microorganism region image sequences of a plurality of assumed damaged microorganisms, wherein each assumed damaged microorganism region image sequence includes a plurality of assumed damaged microorganism region images of the same assumed damaged microorganism.
[0011] In an implementation, the plurality of assumed damaged microorganism region image sequences are subjected to microorganism dynamic recognition to obtain a plurality of damaged microorganism region sets and a plurality of actual stain region sets, the plurality of assumed stain region sets are added to obtain a plurality of stain region sets, including: a sample damaged microorganism region image sequence set is collected, and a sample movement recognition result set is labeled according to whether the damaged microorganism moves. The sample damaged microorganism region image sequence set and the sample movement recognition result set are used to train a damaged microorganism dynamic recognizer. The plurality of assumed damaged microorganism region image sequences are subjected to microorganism dynamic recognition by using the damaged microorganism dynamic recognizer to obtain a plurality of movement recognition results. An assumed damaged microorganism region with a movement recognition result of yes is taken as a damaged microorganism region to obtain a plurality of damaged microorganism region sets, and an assumed damaged microorganism region with a movement recognition result of no is taken as an actual stain region to obtain a plurality of actual stain region sets. The plurality of actual stain region sets are added to the plurality of assumed stain region sets to obtain a plurality of stain region sets. The plurality of damaged microorganism region sets and a plurality of complete microorganism region sets are subjected to microorganism detection.
[0012] The application also provides an image recognition-based microorganism efficient detection device, including: A region acquisition module is configured to collect a microorganism detection image sequence subjected to microorganism detection, perform region of interest recognition on a plurality of microorganism detection images in the microorganism detection image sequence, and obtain a plurality of region of interest sets.
[0013] A key region extraction module is configured to perform complete microorganism recognition on the plurality of region of interest sets to obtain a plurality of complete microorganism region sets, delete the plurality of complete microorganism region sets from the plurality of region of interest sets, and obtain a plurality of key region sets.
[0014] A damage fitting module is configured to perform fitting on pixel points in the plurality of key region sets based on a plurality of damage morphologies of microorganism damage to obtain a plurality of fitting result sets, and screen to obtain a plurality of assumed damaged microorganism region sets and a plurality of assumed stain region sets.
[0015] A clustering module is configured to perform damaged microorganism region clustering on the plurality of assumed damaged microorganism region sets to obtain a plurality of assumed damaged microorganism region image sequences of a plurality of assumed damaged microorganisms.
[0016] A biological detection module is configured to perform microorganism dynamic recognition on the plurality of assumed damaged microorganism region image sequences to obtain a plurality of damaged microorganism region sets and a plurality of actual stain region sets, add the plurality of assumed stain region sets to obtain a plurality of stain region sets, and perform microorganism detection on the plurality of damaged microorganism region sets and a plurality of complete microorganism region sets.
[0017] The microorganism efficient detection method and device based on image recognition provided in the present application can collect a microorganism detection image sequence, identify a region of interest, obtain a plurality of region of interest sets, identify a complete microorganism, and obtain a plurality of complete microorganism region sets. The plurality of complete microorganism region sets in the plurality of region of interest sets are deleted to obtain a plurality of key region sets. The pixels in the plurality of key region sets are fitted to obtain a plurality of assumed damaged microorganism region sets and a plurality of assumed stain region sets. A plurality of assumed damaged microorganism region image sequences of a plurality of assumed damaged microorganisms are obtained through clustering, biological dynamic recognition is performed on the plurality of assumed damaged microorganism region image sequences to obtain a plurality of stain region sets, and the plurality of damaged microorganism region sets and the plurality of complete microorganism region sets are detected. The technical problems that the microorganism detection method in the prior art has low detection efficiency and it is difficult to improve detection accuracy are solved. The damaged microorganisms and the stain regions are efficiently distinguished through the automatic image recognition technology and the deep learning model, the data processing amount of subsequent microorganism detection is significantly reduced, the stain misrecognition is reduced, and the efficiency and accuracy of microorganism detection are improved. BRIEF DESCRIPTION OF DRAWINGS
[0018] In order to more clearly illustrate the technical solutions of the embodiments of the present disclosure, the drawings of the embodiments of the present disclosure will be briefly introduced below. In the present application, a flowchart is used to illustrate the operations performed by the device according to the embodiments of the present application. It should be understood that the foregoing or the following operations are not necessarily performed in sequence. On the contrary, according to the needs, various steps can be processed in reverse order or at the same time. At the same time, other operations can be added to these processes, or a step or several steps of operation can be removed from these processes.
[0019] Figure 1 The flowchart of the microorganism efficient detection method based on image recognition provided by the embodiments of the present application is shown in the figure. Figure 2 The structure diagram of the microorganism efficient detection device based on image recognition provided by the embodiments of the present application is shown in the figure.
[0020] The figure mark explanation: region acquisition module 11, key region extraction module 12, damage fitting module 13, clustering module 14, and biological detection module 15. DETAILED DESCRIPTION
[0021] The above description is only a summary of the technical solutions of the present application. In order to more clearly understand the technical means of the present application, the embodiments of the present application can be implemented according to the content of the specification, and in order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific embodiments of the present application are described below.
[0022] In order to make the purposes, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the drawings. The described embodiments should not be regarded as limitations to the present application. All other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present application.
[0023] In the following description, "some embodiments" are referred to, which describe a subset of all possible embodiments, but it can be understood that "some embodiments" can be the same or different subsets of all possible embodiments, and can be combined with each other without conflict. The term "first\second" referred to is only to distinguish similar objects, and does not represent a specific order of the objects. The terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, device, product or server including a series of steps or units does not have to be limited to those clearly listed steps or units, but can include other steps or modules that are not clearly listed or inherent to these processes, methods, products or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as understood by those skilled in the art to which the present application belongs. The terms used herein are only for the purpose of describing the embodiments of the present application.
[0024] The embodiments of the present application provide a microorganism high-efficiency detection method and device based on image recognition, as shown in the following Figure 1 The method comprises the following steps: Collecting a microorganism detection image sequence for microorganism detection, performing region of interest recognition on a plurality of microorganism detection images in the microorganism detection image sequence, and obtaining a plurality of region of interest sets.
[0025] Performing complete microorganism recognition on the plurality of regions of interest, obtaining a plurality of complete microorganism region sets, deleting the plurality of complete microorganism region sets in the plurality of region of interest sets, and obtaining a plurality of key region sets.
[0026] An image sequence of the microorganism sample is collected using a high-resolution microscope or professional camera equipment. Each image sequence contains several microorganism detection images to ensure coverage of all areas to be detected. The image clarity and consistency should be ensured during the collection process to improve the accuracy of subsequent processing. Subsequently, multiple microorganism detection images in the microorganism detection image sequence are subjected to region of interest identification to obtain multiple region of interest sets, and the region of interest is a region image of a complete microorganism, a damaged microorganism, or a stain. Further, after identifying the multiple regions of interest, further complete microorganism identification is performed on these regions to obtain a multiple complete microorganism region set. Subsequently, the multiple complete microorganism region set is deleted from the multiple region of interest set to obtain a multiple key region set, and each key region contains an image of a damaged microorganism or a stain.
[0027] The method provided by the embodiments of the present application further includes: Based on the historical data of microorganism detection, a sample microorganism detection image set is collected, and a region of interest in each sample microorganism detection image is identified and labeled to obtain a sample region of interest division result set, wherein the region of interest includes an image region of a microorganism and a stain.
[0028] The sample microorganism detection image set and the sample region of interest division result set are used to train a region of interest identifier.
[0029] The region of interest identifier is used to identify a region of interest in multiple microorganism detection images in the microorganism detection image sequence to obtain multiple region of interest sets.
[0030] The process involves collecting a sequence of microbial detection images for microbial testing, identifying regions of interest (ROIs) within multiple microbial detection images in the sequence, and obtaining multiple sets of ROIs. These sets include: a collection of sample microbial detection images containing different types of microorganisms and contaminants, based on historical microbial detection data. These image sets need to cover a wide range of possible microorganisms and contaminant scenarios to ensure the model's generalization ability. 1000 sample images are extracted from past microbial testing projects, including microorganisms and contaminants collected under various environments. ROIs are identified and manually labeled for each sample microbial detection image. This step relies on the experience of professionals to label specific regions including microorganisms and contaminants, forming a set of sample ROI partitioning results. A ROI recognizer is trained using the sample microbial detection image sets and the sample ROI partitioning results sets. This recognizer can be constructed using a deep learning model, such as a convolutional neural network (CNN) or a region convolutional neural network (R-CNN). The model input is the original image, and the output is the image and location of the ROI (including microorganisms and contaminants) for each image. The trained ROI recognizer is then used to identify ROIs within multiple microbial detection images in the microbial detection image sequence. The identifier automatically detects microorganisms and blemishes in the image and generates a set of regions of interest, which contains regional images of multiple regions of interest.
[0031] The method provided in this application also includes: Collect a set of regions of interest for the sample, and label whether the regions of interest include complete microbial images to obtain a set of complete classification information for the sample, where the complete classification information of the sample is 1 or 0.
[0032] Using the sample region of interest set and the sample complete classification information set, a complete microbial region identifier is trained to identify the multiple regions of interest sets and obtain multiple complete classification information sets.
[0033] Extract the regions of interest corresponding to the complete classification information that is 1 in the multiple complete classification information sets, and use them as multiple complete microbial region sets.
[0034] By deleting the multiple complete microbial region sets within the multiple sets of regions of interest, multiple sets of key regions are obtained.
[0035] Based on historical data and newly collected data, a sample set of regions of interest is established, each region of interest is labeled to confirm whether the region contains a complete microorganism image, and a sample complete classification information set is generated. The complete classification information label is 1 or 0, 1 indicating that it contains a complete microorganism, and 0 indicating that it does not. For example: 5000 regions of interest are extracted from 1000 sample images. Each region is manually labeled by a professional to determine whether it contains a complete microorganism, and the corresponding label file is generated. Using the sample set of regions of interest and the sample complete classification information set, a complete microorganism region recognizer is trained. The complete microorganism region recognizer can use a convolutional neural network (CNN) model to identify and classify whether a region of interest contains a complete microorganism. During training, 5000 labeled regions of interest are supervised and trained, with the input being the region of interest image and the output being a classification result of 0 or 1. Using the trained complete microorganism region recognizer, multiple sets of regions of interest detected in practice are identified to generate multiple complete classification information sets. Each region of interest is identified as 0 or 1, with 1 indicating that the region contains a complete microorganism. From the multiple complete classification information sets, the regions of interest with a classification result of 1 are extracted as a complete microorganism region set. This set contains all regions identified as containing a complete microorganism. The complete microorganism region set is deleted from the region of interest set, and the remaining regions are the key region set.
[0036] Based on the various damage morphologies of microorganism damage, the pixel points in the multiple key region sets are fitted to obtain multiple fitting result sets, and multiple assumed damaged microorganism region sets and multiple assumed stain region sets are selected.
[0037] The multiple assumed damaged microorganism region sets are clustered to obtain multiple assumed damaged microorganism region image sequences of multiple assumed damaged microorganisms.
[0038] Microorganism dynamic recognition is performed on the multiple assumed damaged microorganism region image sequences to obtain multiple damaged microorganism region sets and multiple actual stain region sets, which are added to the multiple assumed stain region sets to obtain multiple stain region sets, and microorganism detection is performed on the multiple damaged microorganism region sets and multiple complete microorganism region sets.
[0039] Based on the various damage morphologies of microbial damage, a fitting box is set, the shape and size of the fitting box are similar to the shape and size of the various damage morphologies. Further, the pixel points in the key region set are fitted with the fitting box pixel overlap degree fitting to obtain a plurality of fitting result sets, and the maximum value of the fitting box pixel overlap degree of each key region in the plurality of fitting result sets. Subsequently, the fitting degree threshold is screened to obtain a plurality of assumed damaged microbial region sets and a plurality of assumed stain region sets. Further, the plurality of assumed damaged microbial region sets are clustered to obtain a plurality of assumed damaged microbial region image sequences of the plurality of assumed damaged microorganisms, and each assumed damaged microorganism corresponds to an assumed damaged microbial region image sequence. Since the damaged microorganism will produce displacement in the time sequence, the stain region and the damaged microbial region of the plurality of assumed damaged microbial region image sequences can be distinguished based on whether there is displacement, so as to accurately obtain the damaged microbial region. Specifically, by performing microbial dynamic identification on the plurality of assumed damaged microbial region image sequences, a plurality of damaged microbial region sets and a plurality of actual stain region sets are obtained, the plurality of assumed stain region sets are added to obtain a plurality of stain region sets, and the stain region may be derived from external pollution, rather than the microorganism of the sample itself. By distinguishing these regions, it can be avoided that the stain is mistaken for a microorganism, and false positives or false negatives are prevented. The plurality of damaged microbial region sets and the plurality of complete microbial region sets are subjected to microbial detection. Finally, the plurality of damaged microbial region sets and the plurality of complete microbial region sets are subjected to microbial detection. Through comprehensive analysis, a final detection report is generated, including the state, damage condition and distribution information of each microorganism. A detection report is generated, listing the types and quantities of microorganisms detected, the damage ratio and distribution position of each microorganism, and other detailed information. The technical problems of low detection efficiency and difficult improvement of detection accuracy in the prior art are solved. Through the automatic image recognition technology and the deep learning model, the damaged microorganisms and the stain regions are efficiently distinguished, the data processing amount of subsequent microbial detection is significantly reduced, the stain misrecognition is reduced, and the efficiency and accuracy of microbial detection are improved.
[0040] The method provided by the embodiment of the application further includes: Based on the various damage morphologies of microbial damage, a plurality of fitting boxes are obtained.
[0041] The pixel points in the key region set are iteratively fitted by using the plurality of fitting boxes to obtain a plurality of fitting result sets, wherein each fitting result includes a microbial fitting degree.
[0042] determine whether the microorganism fitting degree in each fitting result is greater than a fitting degree threshold value, if yes, the corresponding key region is regarded as a presumed damaged microorganism region, if no, the corresponding key region is regarded as a presumed stain region, and a plurality of presumed damaged microorganism region sets and a plurality of presumed stain region sets are obtained.
[0043] According to various morphologies (such as rupture, deformation, etc.) of microorganism damage, a plurality of fitting boxes are defined in advance, the shape of the fitting box should be similar to the morphology of the microorganism damage, and the shape and size of the fitting box should be as diverse as possible to cover all possible damage morphologies. The pixel points in the plurality of key region sets are iteratively fitted using the plurality of fitting boxes, and a plurality of fitting result sets are obtained, wherein each fitting result includes a microorganism fitting degree. Further, it is determined whether the microorganism fitting degree in each fitting result is greater than a fitting degree threshold value, if yes, the corresponding key region is regarded as a presumed damaged microorganism region, if no, the corresponding key region is regarded as a presumed stain region, and a plurality of presumed damaged microorganism region sets and a plurality of presumed stain region sets are obtained. The fitting degree threshold value is a fitting degree parameter threshold value set in advance, when greater than the threshold value, the accuracy of the corresponding fitting result is higher, otherwise the fitting result is poor, and the corresponding region is likely to be a stain region.
[0044] The method provided by the embodiment of the application further includes: Randomly select a first fitting box in the plurality of fitting boxes, and perform random box fitting in a first key region in the plurality of key region sets, to obtain a proportion of pixel points in the first key region falling into the first fitting box as a first fitting degree.
[0045] Continue to use other plurality of fitting boxes to fit the first key region, obtain a plurality of fitting degrees, output the maximum fitting degree as a first microorganism fitting degree, and obtain a first fitting result.
[0046] Continue to use the plurality of fitting boxes to iteratively fit the plurality of key region sets, and obtain a plurality of microorganism fitting results as a plurality of fitting result sets.
[0047] The multiple fitting frames are used to iteratively fit the pixel points in the multiple key region sets to obtain multiple fitting result sets, including: randomly selecting a first fitting frame in the multiple fitting frames, the first fitting frame being a random fitting frame. Random frame fitting is performed in a first key region in the multiple key region sets, the first key region being a random key region in the multiple key region sets. The first key region is fitted, and a proportion of pixel points in the key region falling into the first fitting frame is calculated as a first fitting degree. For example, an elliptical fitting frame is randomly placed in a certain key region, and it is calculated that 50% of the pixel points in the region fall into the fitting frame, so the first fitting degree is 0.5. The first key region is fitted using other multiple fitting frames to calculate the fitting degrees, a maximum fitting degree is outputted from the multiple fitting degrees as a first microorganism fitting degree, and a first fitting result is obtained. The same way of iterative fitting is performed on other multiple key regions in the multiple key region sets using the multiple fitting frames to obtain multiple microorganism fitting results as the multiple fitting result sets.
[0048] The method provided in the embodiments of the present application further includes: The assumed damaged microorganism regions in the multiple assumed damaged microorganism region sets are randomly combined, image similarity analysis is performed, and multiple assumed damaged microorganism similarity sets are obtained.
[0049] A preset moving region is obtained, the preset moving region being a region range of a same microorganism moving in the microorganism detection image sequence.
[0050] Assumed damaged microorganism regions in the same preset moving region in the multiple microorganism detection images and having a damaged microorganism similarity greater than a similarity threshold are screened, clustering of the damaged microorganism regions is completed, and sorting is performed according to time sequence, multiple assumed damaged microorganism region image sequences of multiple assumed damaged microorganisms are obtained, and each assumed damaged microorganism region image sequence includes multiple assumed damaged microorganism region images of a same assumed damaged microorganism.
[0051] The clustering of the plurality of assumed damaged microorganism regions is performed on the plurality of assumed damaged microorganism region sets to obtain a plurality of assumed damaged microorganism region image sequences of the plurality of assumed damaged microorganisms, including: randomly combining the assumed damaged microorganism regions in the plurality of assumed damaged microorganism region sets two by two, performing similarity analysis on the randomly combined region images, performing image similarity analysis by using a commonly used similarity analysis method such as structural similarity and Euclidean distance, and obtaining a plurality of assumed damaged microorganism similarity sets. Further, a preset moving region is obtained, which represents the possible moving range of the same microorganism at different time points in the microorganism detection image sequence. Generally, the preset moving region is a rectangular or circular region containing the possible position changes of the microorganism within a certain time. Assuming that the preset moving region is a circular region with a radius of 30 pixels, the moving range of the microorganism in the image sequence is not more than 30 pixels. Finally, the assumed damaged microorganism regions that are in the same preset moving region in the plurality of microorganism detection images and have a damaged microorganism similarity greater than a similarity threshold are screened, and the clustering of the damaged microorganism regions is completed. That is, it is judged whether each pair of damaged microorganism regions is a straight line that is less than or equal to the preset moving region, and the assumed damaged microorganism regions that have a damaged microorganism similarity greater than a similarity threshold are clustered to the same target. The assumed damaged microorganism regions that meet the requirements are taken as a clustering cluster, and an assumed damaged microorganism region image sequence of a target (damaged microorganism or stain) is obtained. The images in the clustering cluster are arranged in time sequence to obtain a plurality of assumed damaged microorganism region image sequences of the plurality of assumed damaged microorganisms, wherein each assumed damaged microorganism region image sequence includes a plurality of assumed damaged microorganism region images of the same assumed damaged microorganism.
[0052] The method provided in the embodiments of the present application further includes: A sample damaged microorganism region image sequence set is collected, and a sample moving identification result set is obtained according to whether the damaged microorganism moves.
[0053] The sample damaged microorganism region image sequence set and the sample moving identification result set are used to train a damaged microorganism dynamic identifier.
[0054] The damaged microorganism dynamic identifier is used to perform microorganism dynamic identification on the plurality of assumed damaged microorganism region image sequences to obtain a plurality of moving identification results.
[0055] The assumed damaged microorganism regions with the moving identification result being yes are taken as damaged microorganism regions to obtain a plurality of damaged microorganism region sets, and the assumed damaged microorganism regions with the moving identification result being no are taken as actual stain regions to obtain a plurality of actual stain region sets.
[0056] The multiple actual stain region sets are added to the multiple assumed stain region sets to obtain multiple stain region sets.
[0057] Microorganism detection is performed on the multiple damaged microorganism region sets and the multiple complete microorganism region sets.
[0058] A sample damaged microorganism region image sequence set is collected, the sample damaged microorganism region image sequence set is obtained based on historical image sequences of different damaged microorganisms, the historical image sequences of the different damaged microorganisms need to cover various possible moving conditions of microorganisms, and each image sequence is labeled with moving objects and non-moving objects to form a sample moving identification result set, the sample moving identification result set contains the moving objects and the non-moving objects. Further, the sample damaged microorganism region image sequence set and the sample moving identification result set are used to train a damaged microorganism dynamic identifier. The damaged microorganism dynamic identifier is constructed based on a recurrent neural network (RNN) model, the sample damaged microorganism region image sequence set and the sample moving identification result set are input into an untrained model for supervised learning, and the model parameters are adjusted to minimize the identification error. The damaged microorganism dynamic identifier is used to perform microorganism dynamic identification on the multiple assumed damaged microorganism region image sequences, the identifier automatically judges whether each assumed damaged microorganism region moves or not, and multiple moving identification results are obtained. The assumed damaged microorganism region with a moving identification result of yes is regarded as a damaged microorganism region to obtain multiple damaged microorganism region sets, and the assumed damaged microorganism region with a moving identification result of no is regarded as an actual stain region to obtain multiple actual stain region sets. The multiple actual stain region sets are added to the multiple assumed stain region sets, that is, the multiple actual stain region sets and the assumed stain region sets are merged to form a complete stain region set, and multiple stain region sets are obtained. Finally, microorganism detection is performed on the multiple damaged microorganism region sets and the multiple complete microorganism region sets.
[0059] In the foregoing, reference is made to Figure 1 The microorganism efficient detection method based on image recognition according to the embodiments of the present application is described in detail. Next, the microorganism efficient detection device based on image recognition according to the embodiments of the present application will be described with reference to Figure 2
[0060] The microorganism efficient detection device based on image recognition according to the embodiment of the present application solves the technical problems of low detection efficiency and difficult improvement of detection accuracy of the microorganism detection method in the prior art. The automatic image recognition technology and the deep learning model are used to efficiently distinguish damaged microorganisms and stain areas, significantly reduce the data processing amount of subsequent microorganism detection, reduce the misrecognition of stains, and improve the efficiency and accuracy of microorganism detection. The microorganism efficient detection device based on image recognition comprises a region acquisition module 11, a key region extraction module 12, a damage fitting module 13, a clustering module 14, and a biological detection module 15.
[0061] The region acquisition module 11 is configured to collect a microorganism detection image sequence for microorganism detection, perform region of interest recognition on a plurality of microorganism detection images in the microorganism detection image sequence, and obtain a plurality of region of interest sets.
[0062] The key region extraction module 12 is configured to perform complete microorganism recognition on the plurality of region of interest sets, obtain a plurality of complete microorganism region sets, delete the plurality of complete microorganism region sets from the plurality of region of interest sets, and obtain a plurality of key region sets.
[0063] The damage fitting module 13 is configured to perform fitting on pixel points in the plurality of key region sets based on a plurality of damage morphologies of microorganism damage, obtain a plurality of fitting result sets, and screen a plurality of assumed damaged microorganism region sets and a plurality of assumed stain region sets.
[0064] The clustering module 14 is configured to perform clustering of damaged microorganism regions on the plurality of assumed damaged microorganism region sets, and obtain a plurality of assumed damaged microorganism region image sequences of a plurality of assumed damaged microorganisms.
[0065] The biological detection module 15 is configured to perform microorganism dynamic recognition on the plurality of assumed damaged microorganism region image sequences, obtain a plurality of damaged microorganism region sets and a plurality of actual stain region sets, add the plurality of assumed stain region sets to obtain a plurality of stain region sets, and perform microorganism detection on the plurality of damaged microorganism region sets and the plurality of complete microorganism region sets.
[0066] In the following, the specific configuration of the region acquisition module 11 will be described in detail. The region acquisition module 11 can further include: collecting a sequence of microorganism detection images for microorganism detection, performing region of interest identification on a plurality of microorganism detection images in the sequence of microorganism detection images, and obtaining a plurality of region of interest sets, including: collecting a set of sample microorganism detection images based on historical data of microorganism detection, performing identification labeling on the region of interest in each sample microorganism detection image, and obtaining a set of sample region of interest division results, wherein the region of interest includes the image region of microorganisms and stains. Using the set of sample microorganism detection images and the set of sample region of interest division results, a region of interest identifier is trained. Using the region of interest identifier, the region of interest identification is performed on a plurality of microorganism detection images in the sequence of microorganism detection images, and a plurality of region of interest sets are obtained.
[0067] In the following, the specific configuration of the key region extraction module 12 will be described in detail. The key region extraction module 12 further includes: deleting the plurality of complete microorganism region sets in the plurality of region of interest sets to obtain a plurality of key region sets, including: collecting a set of sample regions of interest and labeling whether the region of interest includes a complete microorganism image to obtain a set of sample complete classification information, wherein the sample complete classification information is 1 or 0. Using the set of sample regions of interest and the set of sample complete classification information, a complete microorganism region identifier is trained to identify the plurality of region of interest sets to obtain a plurality of complete classification information sets. The region of interest corresponding to the complete classification information of 1 in the plurality of complete classification information sets is extracted as a plurality of complete microorganism region sets. The plurality of complete microorganism region sets are deleted in the plurality of region of interest sets to obtain a plurality of key region sets.
[0068] In the following, the specific configuration of the damage fitting module 13 will be described in detail. The damage fitting module 13 can further include: based on a plurality of damage morphologies of microorganism damage, fitting the pixel points in the plurality of key region sets to obtain a plurality of fitting result sets, and screening to obtain a plurality of presumed damaged microorganism region sets and a plurality of presumed stain region sets, including: based on a plurality of damage morphologies of microorganism damage, obtaining a plurality of fitting boxes. Using the plurality of fitting boxes, the pixel points in the plurality of key region sets are iteratively fitted to obtain a plurality of fitting result sets, wherein each fitting result includes a microorganism fitting degree. It is judged whether the microorganism fitting degree in each fitting result is greater than a fitting degree threshold, if yes, the corresponding key region is taken as a presumed damaged microorganism region, if not, the corresponding key region is taken as a presumed stain region, and a plurality of presumed damaged microorganism region sets and a plurality of presumed stain region sets are obtained.
[0069] The specific configuration of the damage fitting module 13 will be described in detail below. The damage fitting module 13 further comprises: iteratively fitting the pixels in the plurality of key region sets by using the plurality of fitting frames to obtain a plurality of fitting result sets, including: randomly selecting a first fitting frame in the plurality of fitting frames, and performing random frame fitting in the first key region in the plurality of key region sets to obtain the proportion of pixels in the first key region falling into the first fitting frame as a first fitting degree. Continue to use other plurality of fitting frames to fit the first key region to obtain a plurality of fitting degrees, output the maximum fitting degree as the first microorganism fitting degree, and obtain the first fitting result. Continue to use the plurality of fitting frames to iteratively fit the plurality of key region sets to obtain a plurality of microorganism fitting results as the plurality of fitting result sets.
[0070] The specific configuration of the clustering module 14 will be described in detail below. The clustering module 14 further comprises: clustering the plurality of assumed damaged microorganism regions to obtain a plurality of assumed damaged microorganism region image sequences of a plurality of assumed damaged microorganisms, including: randomly combining the assumed damaged microorganism regions in the plurality of assumed damaged microorganism region sets, performing image similarity analysis to obtain a plurality of assumed damaged microorganism similarity sets. Obtain a preset moving region, the preset moving region is the region range of the same microorganism moving in the microorganism detection image sequence. Screen the assumed damaged microorganism regions in the same preset moving region in the plurality of microorganism detection images, and the damaged microorganism similarity is greater than the similarity threshold, complete the clustering of the damaged microorganism region, and sort according to the time sequence to obtain a plurality of assumed damaged microorganism region image sequences of a plurality of assumed damaged microorganisms, wherein each assumed damaged microorganism region image sequence includes a plurality of assumed damaged microorganism region images of the same assumed damaged microorganism.
[0071] The specific configuration of the biological detection module 15 will be described in detail below. The biological detection module 15 further includes: performing microorganism dynamic identification on the plurality of assumed damaged microorganism region image sequences, obtaining a plurality of damaged microorganism region sets and a plurality of actual stain region sets, adding the plurality of assumed stain region sets, and obtaining a plurality of stain region sets, including: collecting a sample damaged microorganism region image sequence set, and labeling a sample movement identification result set according to whether the damaged microorganism moves. Using the sample damaged microorganism region image sequence set and the sample movement identification result set, a damaged microorganism dynamic identifier is trained. Using the damaged microorganism dynamic identifier, microorganism dynamic identification is performed on the plurality of assumed damaged microorganism region image sequences, and a plurality of movement identification results are obtained. An assumed damaged microorganism region with a movement identification result of yes is taken as a damaged microorganism region, and a plurality of damaged microorganism region sets are obtained. An assumed damaged microorganism region with a movement identification result of no is taken as an actual stain region, and a plurality of actual stain region sets are obtained. The plurality of actual stain region sets are added to the plurality of assumed stain region sets, and a plurality of stain region sets are obtained. Microorganism detection is performed on the plurality of damaged microorganism region sets and a plurality of complete microorganism region sets.
[0072] The microorganism efficient detection device based on image recognition provided in the embodiments of the present application can perform the microorganism efficient detection method based on image recognition provided in any of the embodiments of the present application, and has the corresponding function modules and beneficial effects of the execution method.
[0073] Although the present application makes various references to certain modules in the device according to the embodiments of the present application, however, any number of different modules can be used and run on the user terminal and / or server, and the various units and modules are only divided according to the functional logic, but are not limited to the above division, as long as the corresponding functions can be realized; in addition, the specific names of the functional units are only for easy mutual differentiation, and do not limit the protection scope of the present application.
[0074] The specific embodiments described above do not constitute a limitation on the protection scope of the present application. Those skilled in the art should understand that various modifications, combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present application should be included in the protection scope of the present application.
Claims
1. A method for efficient detection of microorganisms based on image recognition, characterized in that, The method comprises: Collecting a microorganism detection image sequence for microorganism detection, performing region of interest identification on a plurality of microorganism detection images in the microorganism detection image sequence to obtain a plurality of region of interest sets; Performing complete microorganism identification on the plurality of regions of interest to obtain a plurality of complete microorganism region sets, deleting the plurality of complete microorganism region sets from the plurality of region of interest sets to obtain a plurality of key region sets; Based on a plurality of damage morphologies of microorganism damage, fitting pixel points in the plurality of key region sets to obtain a plurality of fitting result sets, and screening to obtain a plurality of assumed damaged microorganism region sets and a plurality of assumed stain region sets; Performing damaged microorganism region clustering on the plurality of assumed damaged microorganism region sets to obtain a plurality of assumed damaged microorganism region image sequences of a plurality of assumed damaged microorganisms; Performing microorganism dynamic identification on the plurality of assumed damaged microorganism region image sequences to obtain a plurality of damaged microorganism region sets and a plurality of actual stain region sets, adding the plurality of assumed stain region sets to obtain a plurality of stain region sets, and performing microorganism detection on the plurality of damaged microorganism region sets and the plurality of complete microorganism region sets.
2. The image recognition-based microorganism high-efficiency detection method according to claim 1, characterized in that, Collecting a microorganism detection image sequence for microorganism detection, performing region of interest identification on a plurality of microorganism detection images in the microorganism detection image sequence to obtain a plurality of region of interest sets, comprising: Based on historical data of microorganism detection, collecting a sample microorganism detection image set, identifying and labeling a region of interest in each sample microorganism detection image to obtain a sample region of interest division result set, wherein the region of interest includes image regions of microorganisms and stains; Using the sample microorganism detection image set and the sample region of interest division result set, training a region of interest identifier; Using the region of interest identifier, performing region of interest identification on a plurality of microorganism detection images in the microorganism detection image sequence to obtain a plurality of region of interest sets.
3. The image recognition-based microorganism high-efficiency detection method according to claim 1, characterized in that, Deleting the plurality of complete microorganism region sets from the plurality of region of interest sets to obtain a plurality of key region sets, comprising: Collecting a sample region of interest set and labeling whether a region of interest includes a complete microorganism image to obtain a sample complete classification information set, wherein the sample complete classification information is 1 or 0; Using the sample region of interest set and the sample complete classification information set, training a complete microorganism region identifier to identify the plurality of region of interest sets to obtain a plurality of complete classification information sets; Extracting a region of interest corresponding to complete classification information of 1 in the plurality of complete classification information sets as a plurality of complete microorganism region sets; Deleting the plurality of complete microorganism region sets from the plurality of region of interest sets to obtain a plurality of key region sets.
4. The image recognition-based microorganism high-efficiency detection method according to claim 1, characterized in that, Based on a plurality of damage morphologies of microorganism damage, fitting pixel points in the plurality of key region sets to obtain a plurality of fitting result sets, and screening to obtain a plurality of assumed damaged microorganism region sets and a plurality of assumed stain region sets, comprising: Based on the various damage morphologies of the microbial damage, a plurality of fitting boxes are obtained; The pixel points in the plurality of key region sets are iteratively fitted using the plurality of fitting boxes to obtain a plurality of fitting result sets, wherein each fitting result includes a microbial fitting degree; It is judged whether the microbial fitting degree in each fitting result is greater than a fitting degree threshold. If yes, the corresponding key region is regarded as a suspected damaged microbial region. If no, the corresponding key region is regarded as a suspected stain region. A plurality of suspected damaged microbial region sets and a plurality of suspected stain region sets are obtained.
5. The image recognition-based microorganism high-efficiency detection method according to claim 4, characterized in that, The pixel points in the plurality of key region sets are iteratively fitted using the plurality of fitting boxes to obtain a plurality of fitting result sets, including: A first fitting box is randomly selected from the plurality of fitting boxes, and a random box fitting is performed in a first key region in the plurality of key region sets. The proportion of pixel points in the first key region falling into the first fitting box is obtained as a first fitting degree. Continue to use other fitting boxes to fit the first key region to obtain a plurality of fitting degrees. The maximum fitting degree is output as a first microbial fitting degree, and a first fitting result is obtained. Continue to use the plurality of fitting boxes to iteratively fit the plurality of key region sets to obtain a plurality of microbial fitting results as a plurality of fitting result sets. 6.The image recognition-based microorganism high-efficiency detection method according to claim 1, wherein, The plurality of suspected damaged microbial region sets are clustered to obtain a plurality of suspected damaged microbial region image sequences of a plurality of suspected damaged microorganisms, including: Randomly combine the suspected damaged microbial regions in the plurality of suspected damaged microbial region sets and perform image similarity analysis to obtain a plurality of suspected damaged microbial similarity sets; A preset moving region is obtained, which is the region range of the same microorganism moving in the microbial detection image sequence; Suspected damaged microbial regions in the same preset moving region in the plurality of microbial detection images and having a damaged microbial similarity greater than a similarity threshold are screened. The clustering of damaged microbial regions is completed, and the sorting is performed in time sequence to obtain a plurality of suspected damaged microbial region image sequences of a plurality of suspected damaged microorganisms, wherein each suspected damaged microbial region image sequence includes a plurality of suspected damaged microbial region images of the same suspected damaged microorganism. 7.The image recognition-based microorganism high-efficiency detection method according to claim 1, characterized in that, Microbial dynamic identification is performed on the plurality of suspected damaged microbial region image sequences to obtain a plurality of damaged microbial region sets and a plurality of actual stain region sets. The plurality of suspected stain region sets are added to obtain a plurality of stain region sets, including: A sample damaged microbial region image sequence set is collected, and a sample moving identification result set is labeled according to whether the damaged microorganism moves; The sample damaged microbial region image sequence set and the sample moving identification result set are used to train a damaged microbial dynamic identifier; The damaged microbial dynamic identifier is used to perform microbial dynamic identification on the plurality of suspected damaged microbial region image sequences to obtain a plurality of moving identification results; The assumed damaged microbial regions with the movement recognition result as yes are obtained as damaged microbial regions, a plurality of damaged microbial region sets are obtained, the assumed damaged microbial regions with the movement recognition result as no are obtained as actual stain regions, a plurality of actual stain region sets are obtained; The plurality of actual stain region sets are added to the plurality of assumed stain region sets, and a plurality of stain region sets are obtained; Microbial detection is performed on the plurality of damaged microbial region sets and the plurality of complete microbial region sets.
8. A high-efficiency microorganism detection device based on image recognition, characterized in that, The device comprises: A region acquisition module is configured to collect a microbial detection image sequence for microbial detection, perform region of interest recognition on a plurality of microbial detection images in the microbial detection image sequence, and obtain a plurality of region of interest sets; A key region extraction module is configured to perform complete microbial recognition on the plurality of region of interest sets, obtain a plurality of complete microbial region sets, delete the plurality of complete microbial region sets from the plurality of region of interest sets, and obtain a plurality of key region sets; A damage fitting module is configured to perform fitting on pixel points in the plurality of key region sets based on a plurality of damage morphologies of microbial damage, obtain a plurality of fitting result sets, and screen a plurality of assumed damaged microbial region sets and a plurality of assumed stain region sets; A clustering module is configured to perform clustering on the plurality of assumed damaged microbial region sets to obtain a plurality of assumed damaged microbial region image sequences of a plurality of assumed damaged microorganisms; A biological detection module is configured to perform microbial dynamic recognition on the plurality of assumed damaged microbial region image sequences, obtain a plurality of damaged microbial region sets and a plurality of actual stain region sets, add the plurality of actual stain region sets to the plurality of assumed stain region sets, obtain a plurality of stain region sets, and perform microbial detection on the plurality of damaged microbial region sets and the plurality of complete microbial region sets.