Culture dish microorganism growth monitoring method based on deep learning

Abstract

The invention discloses a culture dish microorganism growth monitoring method based on deep learning, and the method comprises the following steps: (1) obtaining a microorganism image in a normal growth state, and carrying out the preprocessing; (2) constructing an improved Cascade R-CNN (Region Convolutional Neural Network), namely a WY-FPN model; and inputting the image obtained in the step (1) into a WY-FPN model for training to obtain the area and size of the microorganisms and the color of the culture dish solution in a normal growth state. (3) acquiring images of the microorganisms in growth in real time by using a camera acquisition device, and preprocessing the images; (4) inputting the image obtained in the step (3) into a WY-FPN model for training to obtain the area and size of the microorganisms and the color of the culture dish solution in a real-time growth state; (5) comparing the results obtained in the step (4) with the results obtained in the step (2), and evaluating the growth condition of the microorganisms in the real-time growth state; according to the invention, real-time and automatic monitoring of the growth condition of the microbial sample in the culture dish is realized, errors and time cost in the traditional manual observation process are reduced, and the monitoring efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of microbial monitoring, and in particular to a method for monitoring the growth of microorganisms in a culture dish based on deep learning. Background Art

[0002] In traditional microbial growth monitoring, researchers usually use manual observation, recording and analysis to evaluate the growth of microorganisms. Specifically, experimenters need to regularly observe the microbial samples in the culture dish under a microscope, judge the growth degree and distribution of the microorganisms by visual inspection, or indirectly judge the metabolic state of the microorganisms by observing the color changes of the culture medium with the naked eye. In addition, instruments such as photometers are also used to measure turbidity changes in the culture medium as an indirect indicator of microbial growth. However, traditional methods have many limitations, such as relying on manual operation, time-consuming and labor-intensive, and errors may occur due to subjective judgment, making it difficult to carry out large-scale automated monitoring.

[0003] Although traditional manual monitoring methods were widely used in early microbiological research, their limitations became increasingly apparent as the scale and complexity of experiments increased. First, manual observations have low accuracy and cannot provide real-time, continuous monitoring data, making it difficult to meet the data volume and accuracy requirements of modern biological research. Second, traditional equipment such as photometers can only provide rough indirect indicators and cannot accurately reflect the specific growth morphology and distribution of microorganisms, resulting in limitations in the analysis results. Finally, due to reliance on manual records and analysis, traditional methods are inefficient in data storage and processing, making them difficult to automate and manage. Summary of the invention

[0004] Purpose of the invention: The purpose of the present invention is to provide a method for monitoring the growth of microorganisms in a culture dish based on deep learning to solve the problems existing in the background technology.

[0005] Technical solution: The method for monitoring the growth of microorganisms in a culture dish based on deep learning described in the present invention comprises the following steps: (1) Obtain images of microorganisms under normal growth conditions and perform preprocessing; (2) Constructing an improved Cascade R-CNN, namely the WY-FPN model; inputting the image obtained in step (1) into the WY-FPN model for training to obtain the area and size of microorganisms under normal growth conditions and the color of the culture dish solution; (3) Using a video acquisition device to collect images of growing microorganisms in real time and perform preprocessing; (4) Input the image obtained in step (3) into the WY-FPN model for training to obtain the microbial area and size and the color of the culture dish solution in real-time growth state; (5) Compare the results obtained in step (4) with those obtained in step (2) to evaluate the growth of the microorganisms in the real-time growth state.

[0006] Furthermore, in step (1), the microbial image under normal growth state is divided into multiple regions, annotated by labelImage software, and divided into a test set, a training set, and a validation set; the preprocessing includes: eliminating image noise and normalization processing.

[0007] Furthermore, the WY-FPN model in step (2) is specifically as follows: the network structure of Cascade R-CNN is used as the basic architecture, and the feature pyramid module FPN is introduced, including: backbone network, neck network and output layer; wherein, the backbone network includes: convolution layer, multi-scale feature fusion layer MDC, channel attention fusion layer C2f, feature pyramid module FPN; the neck network includes: the neck includes: upsampling, feature splicing, C2f layer and convolution layer.

[0008] Furthermore, the convolution layer is used to capture local features in the image; the multi-scale feature fusion layer is used to integrate features of different scales.

[0009] Furthermore, the channel attention fusion layer C2f identifies key features by introducing advanced attention mechanisms; the feature pyramid network FPN is used to fuse features at different levels.

[0010] Furthermore, upsampling is performed to obtain the resolution of the feature map; feature maps at different levels are integrated through the feature splicing strategy to form a multi-level and multi-dimensional feature set.

[0011] Furthermore, the C2f layer is used to capture the key information of the feature map; the convolutional layer (Conv) fine-tunes the feature map.

[0012] Furthermore, the Focaler-IoU loss function is introduced to optimize the model; the expression of the Focaler-IoU loss function is as follows: in, Expressed as the intersection-over-union ratio of bounding boxes, is the lower threshold, is the upper threshold; when Less than a lower threshold When , the loss is 0; when Greater than an upper threshold When , the loss is 1; and when In and Between, the loss is a A function whose value increases linearly.

[0013] An electronic device described in the present invention includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is loaded into the processor, any one of the methods for monitoring the growth of microorganisms in a culture dish based on deep learning is implemented.

[0014] A storage medium described in the present invention stores a computer program, and when the computer program is executed by a processor, it implements any one of the methods for monitoring the growth of microorganisms in a culture dish based on deep learning.

[0015] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages: Through image processing technology based on deep learning, the present invention can realize real-time and automated monitoring of the growth of microbial samples in culture dishes, reducing the errors and time costs in the traditional manual observation process and improving monitoring efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 is a flow chart of the present invention; Figure 2 It is a model principle diagram of the present invention. DETAILED DESCRIPTION

[0017] The technical solution of the present invention is further described below in conjunction with the accompanying drawings.

[0018] like Figure 1 As shown, an embodiment of the present invention provides a method for monitoring the growth of microorganisms in a culture dish based on deep learning, comprising the following steps: (1) Obtain images of microorganisms under normal growth conditions and perform preprocessing; specifically, divide the images of microorganisms under normal growth conditions into multiple regions, annotate them using labelImage software, and divide them into a test set, a training set, and a validation set; preprocessing includes: eliminating image noise and normalizing the images.

[0019] (2) Construct an improved Cascade R-CNN, namely the WY-FPN model; input the image obtained in step (1) into the WY-FPN model for training to obtain the microbial area and size and the color of the culture dish solution under normal growth conditions. The WY-FPN model is as follows: The network structure of Cascade R-CNN is used as the basic architecture, and the feature pyramid module FPN is introduced, including: a backbone network, a neck network and an output layer; wherein the backbone network includes: a convolution layer, a multi-scale feature fusion layer MDC, a channel attention fusion layer C2f, and a feature pyramid module FPN; the neck network includes: the neck includes: upsampling, feature splicing, C2f layer and convolution layer; the specific process is as follows: In the WY-FPN neural network architecture, the input layer plays a vital role. As the starting point of the entire network, its main task is to receive and process raw image data. In order to ensure that the network can capture every detail in the image, the input layer of the present invention is designed to receive high-resolution images, with a specific size of 608x608 pixels, covering three RGB color channels. Such a design not only improves the precision of image processing, but also lays a solid foundation for the subsequent feature extraction and analysis of the present invention. Through such high-resolution input, the network can more accurately identify and analyze the subtle features in the image, thereby improving the accuracy and reliability of target detection.

[0020] The backbone network of WY-FPN occupies the core position of the network. Its main function is to deeply explore and extract key features in the input image, which have a decisive influence on the realization of accurate target detection. In terms of design, the backbone network cleverly integrates multiple modules to optimize the feature extraction process. It includes a convolutional layer (Conv), which is responsible for capturing local features in the image; a multi-scale feature fusion layer (MDC), which is used to integrate features of different scales, thereby improving the network's recognition ability for targets of various sizes; and a channel attention fusion layer (C2f), which further strengthens the recognition and utilization of key features by introducing advanced attention mechanisms. In order to achieve more efficient feature integration, the backbone network also integrates a feature pyramid network (FPN). The role of FPN is to fuse features at different levels and construct a multi-scale feature map, which provides richer and more detailed information for target detection at different scales of the present invention. Through such a design, the backbone network can not only extract rich features, but also ensure that these features play the greatest role in target detection tasks.

[0021] In the WY-FPN architecture, the neck plays the role of a bridge, connecting the backbone network and the output layer to ensure the smooth transmission and effective use of feature information. Its core mission is to deepen feature processing and fusion, thereby providing the output layer with rich information required for accurate target detection. In terms of design, the neck uses upsampling technology to improve the resolution of feature maps, which significantly enhances the detail expression of feature maps. At the same time, through the feature concatenation strategy, the neck cleverly integrates feature maps of different levels to form a multi-level and multi-dimensional feature set. In addition, the reuse of the C2f layer further enhances the expressiveness of the features, allowing the network to capture key information more keenly. In order to match the dimension of the feature map with the requirements of subsequent detection tasks, the neck also introduces a convolutional layer (Conv). This layer fine-tunes the feature map to ensure that the feature map has the best structure and dimension before being passed to the output layer, thus laying a solid foundation for accurate target detection. Through this series of designs, the neck not only improves the quality and availability of features, but also provides key support for the performance optimization of the entire WY-FPN network.

[0022] As the final link of the neural network, the output layer of WY-FPN is responsible for converting the features extracted and processed by the previous layers into specific target detection results. Its core goal is to achieve accurate recognition and positioning of targets in the image, thereby outputting detailed detection information. At the design level, the output layer generates multi-scale feature maps through a carefully designed detection layer (Detect) to meet the detection needs of targets of different sizes. The sizes of these feature maps are set to 76x76x255, 38x38x255, and 19x19x255, where the dimension 255 covers the number of predicted target categories for each grid unit and the detailed parameters of the bounding box. Such a multi-scale output strategy enables WY-FPN to effectively detect targets of different sizes, thereby significantly improving the accuracy and coverage of the detection process. Through this design, the output layer not only ensures that the network fully captures the target, but also enhances the adaptability and robustness of target detection in complex scenes, providing a strong guarantee for efficient and accurate target detection.

[0023] The present invention uses the Focaler-IoU loss function to analyze and consider the impact of the distribution of difficult samples and simple samples in bounding box regression on the regression results, so that the detection box fits the detected object more closely. The expression of the Focaler-IoU loss function is as follows: ; in, Expressed as the intersection-over-union ratio of bounding boxes, is the lower threshold, is the upper threshold; when Less than a lower threshold When , the loss is 0; when Greater than an upper threshold When , the loss is 1; and when In and Between, the loss is a A function whose value increases linearly.

[0024] (3) Using a video acquisition device to collect images of growing microorganisms in real time and perform preprocessing; the specific process is the same as step (1); (4) The image obtained in step (3) is input into the WY-FPN model for training to obtain the microbial area and size and the color of the culture dish solution in the real-time growth state; the specific training process is the same as step (2).

[0025] (5) Compare the results obtained in step (4) with those obtained in step (2) to evaluate the growth of the microorganisms in the real-time growth state.

[0026] An embodiment of the present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the computer program is loaded into the processor, the computer program implements any one of the methods for monitoring the growth of microorganisms in a culture dish based on deep learning.

[0027] An embodiment of the present invention further provides a storage medium storing a computer program, wherein the computer program, when executed by a processor, implements any one of the methods for monitoring the growth of microorganisms in a culture dish based on deep learning.

Claims

1. A method for monitoring the growth of microorganisms in a culture dish based on deep learning, characterized in that: The following steps are involved: (1) Obtain images of microorganisms under normal growth conditions and perform preprocessing; (2) Constructing an improved Cascade R-CNN, namely the WY-FPN model; inputting the image obtained in step (1) into the WY-FPN model for training to obtain the area and size of microorganisms under normal growth conditions and the color of the culture dish solution; (3) Using a video acquisition device to collect images of growing microorganisms in real time and perform preprocessing; (4) Input the image obtained in step (3) into the WY-FPN model for training to obtain the microbial area and size and the color of the culture dish solution in real-time growth state; (5) Compare the results obtained in step (4) with those obtained in step (2) to evaluate the growth of the microorganisms in the real-time growth state.

2. A method for monitoring the growth of microorganisms in a culture dish based on deep learning according to claim 1, characterized in that: In step (1), the microbial image under normal growth state is divided into multiple regions, annotated by labelImage software, and divided into a test set, a training set, and a validation set; the preprocessing includes: eliminating image noise and normalization processing.

3. The method for monitoring the growth of microorganisms in a culture dish based on deep learning according to claim 1, characterized in that: Step (2) The WY-FPN model is as follows: The network structure of Cascade R-CNN is used as the basic architecture. The feature pyramid module FPN is introduced, including: a backbone network, a neck network and an output layer; wherein, the backbone network includes: a convolutional layer, a multi-scale feature fusion layer MDC, a channel attention fusion layer C2f, and a feature pyramid module FPN; the neck network includes: the neck includes: upsampling, feature splicing, C2f layer and convolutional layer.

4. A method for monitoring the growth of microorganisms in a culture dish based on deep learning according to claim 3, characterized in that: Convolutional layers are used to capture local features in images; The multi-scale feature fusion layer is used to integrate features of different scales.

5. The method for monitoring the growth of microorganisms in a culture dish based on deep learning according to claim 3, characterized in that: The channel attention fusion layer C2f identifies key features by introducing advanced attention mechanisms; Feature Pyramid Network (FPN) is used to fuse features at different levels.

6. The method for monitoring the growth of microorganisms in a culture dish based on deep learning according to claim 3, characterized in that: Upsampling Upsample obtains the resolution of the feature map; feature maps at different levels are integrated through the feature splicing strategy to form a multi-level, multi-dimensional feature set.

7. The method for monitoring the growth of microorganisms in a culture dish based on deep learning according to claim 3, characterized in that: The C2f layer is used to capture the key information of the feature map; the convolution layer (Conv) fine-tunes the feature map.

8. The method for monitoring the growth of microorganisms in a culture dish based on deep learning according to claim 3, characterized in that: The Focaler-IoU loss function is introduced to optimize the model; the expression of the Focaler-IoU loss function is as follows: ； in, Expressed as the intersection-over-union ratio of bounding boxes, is the lower threshold, is the upper threshold; when Less than a lower threshold When , the loss is 0; when Greater than an upper threshold When , the loss is 1; and when In and Between, the loss is a A function whose value increases linearly.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the computer program is loaded into a processor, a method for monitoring the growth of microorganisms in a culture dish based on deep learning is implemented according to any one of claims 1 to 8.

10. A storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, a method for monitoring the growth of microorganisms in a culture dish based on deep learning is implemented according to any one of claims 1 to 8.

Images