CN-121982391-A - Method, device and storage medium for intelligent analysis of bacterial colony based on deep learning

CN121982391ACN 121982391 ACN121982391 ACN 121982391ACN-121982391-A

Abstract

The invention relates to a colony intelligent analysis method, a device and a storage medium based on deep learning, which are applied to the technical field of microorganism detection and image analysis, and comprise the steps of carrying out model optimization aiming at the characteristics of colony images based on YOLOv deep learning models and combining a migration learning method, remarkably improving the precision and robustness of colony detection, effectively solving the problems of missed detection and false detection caused by overlapping of colonies, multiple forms and complex background, supporting simultaneous identification and classification statistics of multiple types of colonies, automatically generating a structured Excel report containing abundant information, automatically generating the structured Excel report without manual secondary data arrangement, improving the standardization and traceability of colony analysis, providing reliable data support for experimental study and detection evaluation, realizing the full-flow automation of colony detection, classification, counting and report generation, greatly reducing the labor intensity of staff, improving the colony analysis efficiency and meeting the requirement of high-throughput detection of the colonies.

Inventors

ZHU WENJING
LIU QINGQUAN
LIU SHUHUA
YUE HUIZHEN
LI XUEYAN
Tian Jinhao

Assignees

首都医科大学附属北京中医医院

Dates

Publication Date: 20260505
Application Date: 20260120

Claims (10)

1. The intelligent analysis method for the bacterial colony based on the deep learning is characterized by comprising the following steps of: Acquiring colony images covering different forms, backgrounds and categories, marking the colony images, wherein marking information comprises the boundary box position of each colony target and the category of the category, and forming a colony image data set with marking; Optimizing training on the labeled colony image data set by utilizing a migration learning method based on a pre-trained YOLOv model architecture, minimizing a preset overall loss function by using a back propagation algorithm to optimize model parameters until a preset training stopping condition is met, and obtaining a YOLOv model after training is completed; Inputting a colony image to be analyzed into the YOLOv model which is trained, automatically performing forward reasoning by the model, and outputting a prediction boundary frame, belonging class and confidence of each colony target in the colony image to be analyzed; And carrying out automatic statistics on the final colony detection and classification results, calculating the number of different types of colonies in each colony image to be analyzed, carrying out structural integration on the number statistics result and corresponding image identification information, and automatically generating an Excel format report containing the colony types, the number, the image number, the detection time and the confidence threshold.
2. The method of claim 1, wherein the step of determining the position of the substrate comprises, The preset overall loss function includes: Obtaining a bounding box regression loss, a target confidence loss and a classification loss; respectively setting weight coefficients of the regression loss of the bounding box, the target confidence loss and the classification loss; and carrying out weighted summation on the boundary box regression loss, the target confidence coefficient loss and the classification loss according to the weight coefficient to obtain the preset overall loss function.
3. The method as recited in claim 2, further comprising: The model performance verification comprises the step of adopting an average precision mean value, an accuracy rate, a recall rate and an F1 score as evaluation indexes of the model performance.
4. A method according to claim 3, further comprising: image enhancement processing is performed on the annotated colony image dataset, including but not limited to one or more of random cropping, flipping, rotation, brightness adjustment, contrast adjustment, gaussian noise addition, and blurring processing of the image.
5. The method of claim 4, wherein the step of determining the position of the first electrode is performed, The step of structurally integrating the number statistics with the corresponding image identification information further comprises: And storing the prediction boundary frame coordinates and the confidence value of each colony target and the corresponding colony class and quantity information in a correlated manner, wherein the Excel format report supports custom field configuration and data export.
6. The method as recited in claim 5, further comprising: the iterative updating of the model comprises the steps of periodically collecting new colony image data, supplementing the colony image data set and re-executing the model training step, and carrying out iterative optimization on the trained model so as to adapt to new colony morphology and class identification requirements.
7. The method of claim 6, wherein the step of providing the first layer comprises, The preset training stop condition is that the preset total loss function value is not reduced any more, or a preset iteration number is reached.
8. Intelligent analysis device of bacterial colony based on degree of depth study, its characterized in that, the device includes: the training data set acquisition module is used for acquiring colony images covering different forms, backgrounds and categories, labeling the colony images, wherein labeling information comprises the boundary frame position of each colony target and the category of the belonging category, and forming a colony image data set with labeling; The model training module is used for carrying out optimization training on the colony image dataset with the label by utilizing a migration learning method based on a pre-trained YOLOv model architecture, minimizing a preset overall loss function by using a back propagation algorithm to optimize model parameters until a preset training stopping condition is met, and obtaining a YOLOv model after training is completed; The image detection module is used for inputting the colony image to be analyzed into the YOLOv model which is trained, automatically performing forward reasoning by the model, and outputting a prediction boundary frame, belonging class and confidence of each colony target in the colony image to be analyzed; and the report generation module is used for carrying out automatic statistics on the final colony detection and classification results, calculating the number of different types of colonies in each colony image to be analyzed, carrying out structural integration on the number statistics result and corresponding image identification information, and automatically generating an Excel format report containing colony types, number, image numbers, detection time and confidence threshold.
9. The apparatus as recited in claim 8, further comprising: The model verification module is used for verifying the model performance by adopting the average precision mean value, the precision rate, the recall rate and the F1 fraction.
10. A storage medium storing a computer program which, when executed by a master, performs the steps of the deep learning-based intelligent colony analysis method of any of claims 1-7.

Description

Method, device and storage medium for intelligent analysis of bacterial colony based on deep learning Technical Field The invention relates to the technical field of microorganism detection and image analysis, in particular to a colony intelligent analysis method and device based on deep learning and a storage medium. Background In the fields of microbiological experiments, environmental monitoring, food safety, medical health and the like, counting and classification of bacterial colonies are core analysis links, and results directly influence experimental conclusion, detection evaluation and decision making. The traditional colony analysis relies on manual visual interpretation, workers need to count and classify colonies one by one under a microscope or a culture dish, and the method has the remarkable defects of extremely low efficiency, difficulty in meeting the high-flux detection requirement, strong subjectivity, difference in interpretation standards of different workers, easiness in counting errors and classification errors caused by visual fatigue, high labor intensity and easiness in causing fatigue of the workers after long-term repeated work. With the development of image processing technology, an automatic counting method based on traditional machine vision such as threshold segmentation and edge detection is developed. However, the method has poor adaptability to image background and colony morphology, and under the scene of dense colony distribution, multiple morphologies, overlapping or complex background, the problems of missing detection and false detection are easy to occur, the recognition accuracy and the robustness are insufficient, and the requirement of accurate analysis is difficult to meet. In recent years, a target detection algorithm based on deep learning has made breakthrough progress in the field of image recognition by virtue of its strong feature extraction capability. The YOLO series model has high detection speed and high precision, and is widely applied to a plurality of fields such as industrial detection, medical image analysis and the like. However, the existing general YOLO model is not optimized for special targets with tiny forms, tiny differences among classes and dense distribution, and the problems of inaccurate positioning, confusion classification, poor recognition effect of overlapped colonies and the like still exist when the model is directly applied to colony analysis. In addition, most of the existing deep learning colony analysis methods can only realize a single counting function, lack multi-class classification statistical capability, cannot automatically generate a structured analysis report, need manual secondary data arrangement, are difficult to support a complete automatic analysis flow, and limit the application value in actual scenes. Therefore, a full-flow intelligent analysis scheme capable of realizing accurate positioning of colonies, multi-class classification, automatic counting and structured report generation is needed, so as to solve the problems of low efficiency, insufficient precision and single function in the prior art. Disclosure of Invention In view of the above, the invention aims to provide a colony intelligent analysis method, a colony intelligent analysis device and a storage medium based on deep learning, which aim to solve the problems of low colony analysis efficiency, poor precision, single function and insufficient degree of automation in the prior art. According to a first aspect of embodiments of the present invention, there is provided a method for intelligent analysis of colonies based on deep learning, the method comprising: Acquiring colony images covering different forms, backgrounds and categories, marking the colony images, wherein marking information comprises the boundary box position of each colony target and the category of the category, and forming a colony image data set with marking; Optimizing training on the labeled colony image data set by utilizing a migration learning method based on a pre-trained YOLOv model architecture, minimizing a preset overall loss function by using a back propagation algorithm to optimize model parameters until a preset training stopping condition is met, and obtaining a YOLOv model after training is completed; Inputting a colony image to be analyzed into the YOLOv model which is trained, automatically performing forward reasoning by the model, and outputting a prediction boundary frame, belonging class and confidence of each colony target in the colony image to be analyzed; And carrying out automatic statistics on the final colony detection and classification results, calculating the number of different types of colonies in each colony image to be analyzed, carrying out structural integration on the number statistics result and corresponding image identification information, and automatically generating an Excel format report containing the colony types, the number, the imag