CN-121981991-A - Flat-plate colony automatic analysis and counting method and device based on deep learning and electronic equipment

CN121981991ACN 121981991 ACN121981991 ACN 121981991ACN-121981991-A

Abstract

The application provides a method and a device for automatically analyzing and counting plate colonies based on deep learning and electronic equipment. The method comprises the steps of shooting an original image containing a culture dish through a mobile terminal, carrying out culture dish target detection on the original image by utilizing a first target detection model, judging whether the image is a qualified image according to a detection result, classifying the qualified image by utilizing a classification model, classifying the qualified image into three types of a sterile colony image, a low-density colony image and a high-density colony image according to the distribution state of colonies in the image, carrying out colony target detection on the image classified into the low-density colony image by utilizing a second target detection model, identifying and positioning each colony in the image, counting the number of colonies based on the detection result, and uploading the number of the colonies to a data processing system for recording and statistical analysis. And the detection efficiency is remarkably improved from image acquisition, culture dish positioning, abnormal screening and density classification to full-flow automation of colony detection and counting.

Inventors

QIN YINGLIN
YANG BIN
LI YANPENG
NIU MIN
HU YIYONG
HUANG DANCHENG
Zhang gun

Assignees

牧原食品股份有限公司

Dates

Publication Date: 20260505
Application Date: 20260122

Claims (10)

1. An automatic plate colony analysis and counting method based on deep learning is characterized by comprising the following steps: Shooting and obtaining an original image containing the culture dish through the mobile terminal; Performing petri dish target detection on the original image by using a first target detection model, judging whether the image is a qualified image according to a detection result, prompting to re-shoot if the image is a disqualified image, and entering a subsequent step if the image is a qualified image; Classifying the qualified images by using a classification model, and dividing the qualified images into three types of sterile colony images, low-density colony images and high-density colony images according to the distribution states of the colonies in the images; for the image classified into the low-density colony image, performing colony target detection by using a second target detection model, identifying and positioning each colony in the image, and counting the number of colonies based on the detection result; and uploading the colony number to a data processing system for recording and statistical analysis.
2. The method of claim 1, wherein performing dish object detection on the original image using a first object detection model, and determining whether the image is a qualified image based on the detection result comprises: Labeling a culture dish area in the original image, and constructing a culture dish detection training data set; training a first target detection model based on a convolutional neural network using the training data set; and (3) reasoning the input original image by using the trained first target detection model, judging the original image as a qualified image if only one culture dish target frame meeting the preset size and position conditions is detected, and judging the original image as a disqualified image if the culture dish target frame is detected.
3. The method of claim 2, wherein the first object detection model employs an end-to-end object detection network architecture, comprising a feature extraction module, a region suggestion module, and a detection output module.
4. The method of claim 1, wherein classifying the qualified images using a classification model according to a distribution state of colonies in an image into three categories of a sterile colony image, a low-density colony image, and a high-density colony image comprises: dividing qualified images into three categories according to the existence of colonies, the aggregation degree of the colonies and the distribution density of the colonies; Labeling each kind of image and constructing a colony classification training data set; Training a classification model based on a convolutional neural network using the classification training data set, the classification model outputting probabilities that an image belongs to one of the three categories.
5. The method of claim 4, wherein the number of colonies of the low density colony image does not exceed an image of a preset threshold value and the number of colonies of the high density colony image exceeds an image of the preset threshold value.
6. The method of claim 1, wherein the colony detection and counting comprises: Labeling colonies in the low-density colony image, and constructing a colony detection training data set; training a second target detection model based on a convolutional neural network using the training data set; reasoning the low-density colony image by using the trained second target detection model, and outputting colony target frame information; and counting the number based on the colony target frame information.
7. The method of claim 6, further comprising performing post-processing operations on the detected colony target boxes prior to counting the number of colonies, the post-processing operations including non-maximal suppression and overlap box merging.
8. The method of claim 1, wherein the first object detection model and the second object detection model employ the same object detection network architecture and are independently trained via different training data.
9. An automatic plate colony analysis counting device based on deep learning, which is characterized by comprising: The acquisition module is used for shooting and acquiring an original image containing the culture dish through the mobile terminal; The detection module is used for carrying out petri dish target detection on the original image by utilizing a first target detection model, judging whether the image is a qualified image or not according to a detection result, prompting to re-shoot if the image is a disqualified image, and entering a subsequent step if the image is a qualified image; The classification module is used for classifying the qualified images by using a classification model, and classifying the qualified images into three types of sterile colony images, low-density colony images and high-density colony images according to the distribution states of the colonies in the images; The counting module is used for carrying out colony target detection on the image classified into the low-density colony image by utilizing the second target detection model, identifying and positioning each colony in the image, and counting the number of the colonies based on the detection result; And the recording module is used for uploading the colony number to the data processing system for recording and statistical analysis.
10. An electronic device comprising a processor and a memory, the memory storing machine executable instructions executable by the processor to implement the method of any one of claims 1 to 8.

Description

Flat-plate colony automatic analysis and counting method and device based on deep learning and electronic equipment Technical Field The application relates to the technical field of bacterial colonies, in particular to a method and a device for automatically analyzing and counting flat bacterial colonies based on deep learning and electronic equipment. Background In the whole chain from pig raising to slaughtering, a laboratory is required to monitor microorganisms in pigs and the environment in real time, wherein the detection of the number of the microorganisms is a very important link, and the number of the microorganisms not only reflects the health condition of a pig group, but also can be used for confirming whether the pig group has certain diseases in the growing process, and can be used for timely human intervention and medication so as to ensure that the pig group grows healthily at the fastest growing speed. Most laboratories still use a way to evaluate manual counts for statistics. Laboratory personnel evaluate the level of spread and the judgment is greatly affected by subjective factors. Has great influence on the accuracy of the health integrity evaluation of the pig herd. Disclosure of Invention The embodiment of the application aims to provide a method and a device for automatically analyzing and counting plate colonies based on deep learning and electronic equipment, which are used for solving the problems. In a first aspect, the present invention provides a method for automatically analyzing and counting plate colonies based on deep learning, comprising: Shooting and obtaining an original image containing the culture dish through the mobile terminal; Performing petri dish target detection on the original image by using a first target detection model, judging whether the image is a qualified image according to a detection result, prompting to re-shoot if the image is a disqualified image, and entering a subsequent step if the image is a qualified image; Classifying the qualified images by using a classification model, and dividing the qualified images into three types of sterile colony images, low-density colony images and high-density colony images according to the distribution states of the colonies in the images; for the image classified into the low-density colony image, performing colony target detection by using a second target detection model, identifying and positioning each colony in the image, and counting the number of colonies based on the detection result; and uploading the colony number to a data processing system for recording and statistical analysis. In an alternative embodiment, performing dish target detection on the original image by using a first target detection model, and determining whether the image is a qualified image according to a detection result includes: Labeling a culture dish area in the original image, and constructing a culture dish detection training data set; training a first target detection model based on a convolutional neural network using the training data set; and (3) reasoning the input original image by using the trained first target detection model, judging the original image as a qualified image if only one culture dish target frame meeting the preset size and position conditions is detected, and judging the original image as a disqualified image if the culture dish target frame is detected. In an alternative embodiment, the first object detection model adopts an end-to-end object detection network structure, and comprises a feature extraction module, a region suggestion module and a detection output module. In an alternative embodiment, the classifying the qualified image by using a classification model, classifying the qualified image into three categories of a sterile colony image, a low-density colony image and a high-density colony image according to the distribution state of colonies in the image comprises: dividing qualified images into three categories according to the existence of colonies, the aggregation degree of the colonies and the distribution density of the colonies; Labeling each kind of image and constructing a colony classification training data set; Training a classification model based on a convolutional neural network using the classification training data set, the classification model outputting probabilities that an image belongs to one of the three categories. In an alternative embodiment, the number of colonies of the low-density colony image does not exceed an image of a preset threshold value, and the number of colonies of the high-density colony image exceeds an image of the preset threshold value. In an alternative embodiment, the colony detection and enumeration comprises: Labeling colonies in the low-density colony image, and constructing a colony detection training data set; training a second target detection model based on a convolutional neural network using the training data set; reasoning the low-density c