Intelligent algae analysis method and system

Abstract

The invention provides an intelligent algae analysis method and system, and the method comprises the steps: S1, collecting a microscopic image in a microfluidic channel through a flow analyzer based on a target water body; s2, accurately marking the category, the position and the boundary characteristics of the microalgae in the microscopic image by using a marking tool supporting image segmentation and classification in combination with corresponding microalgae classification data; s3, based on the pre-training weight of the Ground SAM, performing fine tuning optimization on the model by using a transfer learning method so as to adapt to microalgae detection; s4, preprocessing the microscopic image acquired in the step S1; s5, for the preprocessed microscopic image, on the basis of the training model in the step S3, realizing target detection and segmentation of microalgae in the microscopic image by using Ground SAM; and S6, counting the types and volumes of the microalgae, identifying the types of the algae, and calculating the density of the algae. Under the support of the flow analyzer, each microalgae cell is accurately detected and counted, so that more detailed algae density data is obtained.

Description

Technical Field

[0001] This application relates to the technical field of water quality monitoring. Specifically, it relates to an intelligent algae analysis method and system. Background Art

[0002] Algal density is one of the important indicators to measure the health status of water bodies. Especially in eutrophic water bodies where algal blooms occur frequently, accurately assessing algal density is crucial for water quality management and ecological protection. In recent years, with the intensification of the global water eutrophication problem, the outbreak frequency of cyanobacterial blooms has been increasing continuously, especially in important lake areas such as Taihu Lake, Chaohu Lake, and Dianchi Lake in China. These algal blooms not only seriously affect the dissolved oxygen concentration and water quality of water bodies, but also lead to the deterioration of the aquatic ecosystem, and even threaten public health and drinking water safety.

[0003] Monitoring algal density is not only of great significance for evaluating the degree of water eutrophication and predicting the timing of algal bloom outbreaks, but also plays a crucial role in aspects such as water quality protection, ecological restoration, and water resource management. For example, in eutrophic water bodies, a higher algal density often means a potential risk of algal bloom outbreaks, and at this time, taking effective water quality control measures is particularly critical. Through accurate algal density estimation, more accurate data support can be provided for water quality managers to take timely measures to control algal growth and avoid the negative impacts brought by algal blooms.

[0004] Traditional methods for monitoring algal density mostly rely on technologies such as manual microscope observation and flow cytometer analysis. Although these methods can provide information about algal species and density to a certain extent, they have many limitations. Traditional microscope methods require a large amount of manual operation, are time-consuming and prone to human errors, and are difficult to achieve large-scale and rapid monitoring; although flow cytometers can provide a relatively high processing speed, their accuracy is still insufficient to cope with the subtle differences between different algal species, and their resolution of morphological characteristics is limited, making it impossible to effectively conduct in-depth algal classification and density analysis. Summary of the Invention

[0005] In view of this, this application provides an intelligent algae analysis method and system to obtain more detailed algal density data and significantly improve the detection accuracy.

[0006] To achieve the above object, the technical solution adopted in this application is as follows: An intelligent algae analysis method, comprising: S1: Based on the target water body, use a flow analyzer to collect microscopic images in a microfluidic channel; S2: Use an annotation tool that supports image segmentation and classification, and combine relevant data on microalgae classification to accurately annotate the category, location, and boundary features of microalgae in microscopic images. Then, verify and optimize the annotated data to provide a high-quality annotated dataset for subsequent model training. S3: Based on the pre-trained weights of Grounded SAM, use transfer learning to fine-tune and optimize the model to adapt to microalgae detection. S4: Preprocess the microscopic images collected in step S1. That is, after performing noise reduction using bilateral filtering, apply the Laplace operator to extract edge information from the background and enhance the boundary features of microalgae. S5: For the preprocessed microscopic images, use the Grounded SAM trained in step S3 to achieve object detection and segmentation of microalgae in microscopic images, obtain the mask map corresponding to the microalgae, and segment the microalgae in the image through the mask map. S6: Count the types and volumes of microalgae, identify the types of algae, and calculate the algae density.

[0007] Furthermore, the algae intelligent analysis method further includes: S7: Feed back the calculated algae density result to the user through a communication protocol.

[0008] Furthermore, the algae intelligent analysis method further includes: The specific steps of step S3 are as follows: S301: Format the annotated dataset in step S2 to meet the input requirements of the Grounded SAM model. S302: Select a widely trained Grounded SAM pre-trained model and adapt the trained model specifically to the microalgae detection task by fine-tuning the high-level features and output layer of the model. S303: During the training process, monitor the performance of the model through loss values and accuracy metrics, and adjust the hyperparameters of the model according to these metrics. The hyperparameters include learning rate, batch size, number of training epochs, and weights of the loss function to ensure the convergence and accuracy of model training. S304: After training is completed, further optimize the model. The optimization methods include pruning and quantization to improve the inference efficiency, save the weights of the fine-tuned model, and evaluate the model using the test set. The evaluation metrics include accuracy, recall, and F1 score.

[0009] Furthermore, the specific steps of step S4 are as follows: S401: Convert the microscopic images collected in step S1, that is, RGB images into grayscale images ; S402: Apply bilateral filtering to the grayscale image Perform noise reduction while preserving the edge information of the image. The specific formula is: where is the spatial distance weight, controlled by to control the smoothing range, is the pixel similarity weight, controls the similarity threshold, is the coordinate of the current processing center pixel, are the coordinates of other pixels in the neighborhood; S403: Use the Laplacian operator to perform a convolution operation on the denoised image to extract edge information. The specific formula is: , where is the Laplacian convolution kernel; S404: Superimpose the extracted edge information on the original image to generate an enhanced edge image , where is the edge enhancement weight coefficient.

[0010] Furthermore, the specific steps of S5 are as follows: S501: Use the image enhanced by the Laplacian operator as the input and pass it to the Grounded SAM model fine-tuned and optimized in step S3 for inference; S502: Perform object detection on the input image through the Grounded SAM model to generate the bounding boxes of microalgae, extract the regions of interest, and output the class label set and the bounding box set , where , representing the upper left and lower right coordinates of each bounding box respectively; S503: Based on the feature extraction and classification of the target region, predict each pixel point in the enhanced image to generate the corresponding microalgae mask , where, , where is a binary matrix; S504: Apply morphological operations to the segmentation mask of each target region to remove pseudo-targets and noise points, optimize the segmentation boundary, and the processed segmentation mask set is: ; S505: Map each optimized target region mask back to the original coordinate system of the enhanced global image to generate the complete pixel-level global segmentation mask , where , and the global segmentation mask Label the pixel categories of all target regions in the microscopic image; S506: Multiply the enhanced original image by the global segmentation mask pixel by pixel to generate an image containing only the microalgae region , and .

[0011] Further, the specific steps of step S6 are as follows: S601: Count the number of targets of each type of microalgae in the image containing only the microalgae region generated in step S506 and the target detection results, and accumulate according to the pixel points belonging to the category in the segmentation mask: , where represents the target belonging to the category ; ; S602: Estimate the volume of each target according to the pixel area of each segmentation mask and the magnification of the microscope: : where is the total number of mask pixel points, is the actual physical area corresponding to the unit pixel area determined by the microscope parameters, and summarize the volumes of all targets by category ; ; S603: Calculate the number of microalgae in category according to the preset unit volume of microalgae in the experiment , and , and the preset unit volume of microalgae is a fixed value determined by measuring a standard sample; ; S604: Calculate the density of category according to the sampling volume of the visible field of view of the microscope and the number of target types : The total density of all targets is calculated as: .

[0012] An intelligent algae analysis system, which is used to implement an intelligent algae analysis method described in this application. The system includes a flow cytometer, an edge computing device, and a cloud server; the flow cytometer is connected to the edge computing device; the flow cytometer is configured with a flow cytometry microscopy device, and the flow cytometry microscopy device is used to capture high-resolution microscopic images of sub-visible and visible microalgae in the liquid flow of the microfluidic channel and upload the microscopic images to the edge computing device according to the sampling period; the edge computing device is used to perform target detection and segmentation processing on the microscopic images, that is, RGB images, at the edge; the edge computing device communicates with the cloud server through a network; the cloud server is used to identify algae and calculate the algae density.

[0013] Compared with the prior art, the beneficial effects of this application are as follows: 0. With the support of the flow cytometer, this application accurately detects and counts each microalgae cell, so as to obtain more detailed algae density data; 1. The system combines edge computing and the Grounded SAM model to achieve real-time detection and segmentation of microalgae targets, which can significantly improve the detection accuracy and play a greater role in real-time monitoring and early warning, with the characteristics of high efficiency, intelligence, and automation; 2. The system can accurately identify the types of microalgae, generate pixel-level segmentation results, greatly improve the detection accuracy and efficiency, and reduce manual intervention; 3. Provide accurate data on algae density and distribution, provide scientific support for early warning of cyanobacterial blooms and water quality management, and contribute to ecological protection and resource management. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the technical solutions of the embodiments of this application, the following will briefly introduce the drawings required in the embodiments. It should be understood that the following drawings only show some embodiments of this application, and therefore should not be regarded as limiting the scope. For those of ordinary skill in the art, without creative efforts, other related drawings can also be obtained based on these drawings.

[0015] Figure 1 It is a flowchart of an intelligent algae analysis method of this application; Figure 2 It is another flowchart of an intelligent algae analysis method of this application; Figure 3 It is a system architecture diagram of an intelligent algae analysis system of this application; Figure 4 It is a structural block diagram of an intelligent algae analysis system of this application; Figure 5 It is a binary mask map of the water area in the specific implementation manner of this application; Figure 6 This is the recognition result diagram of Aphanizomenon flos-aquae in the specific implementation manner of this application. Specific implementation manner

[0016] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of this application. Obviously, the described embodiments are some, but not all, of the embodiments of this application.

[0017] In recent years, with the continuous progress of artificial intelligence and image processing technologies, the method for estimating algal density based on flow cytometry microscopy images has gradually become a new research direction. This method can accurately detect and count each microalgae cell with the support of a flow cytometer through high-resolution microscopy imaging technology, thereby obtaining more detailed algal density data. Especially when combined with deep learning technologies (such as image segmentation, object detection, and classification), the detection accuracy can be significantly improved, and it can play a greater role in real-time monitoring and early warning.

[0018] As Figures 1 - 3 shown, this application provides an intelligent algal analysis method, including: S1: Based on the target water body, use a flow cytometer to collect microscopy images in a microfluidic channel; Specifically, the flow cytometer collects microscopy images according to the sampling period. The collected data must meet the requirements of subsequent model training and target detection. To improve the segmentation and recognition accuracy, the image data should include microalgae samples under various environmental backgrounds, such as microalgae of different species, different sizes, and background complexities.

[0019] S2: Use an annotation tool that supports image segmentation and classification, and accurately annotate the category, location, and boundary features of microalgae in the microscopy images in combination with relevant microalgae classification materials, and verify and optimize the annotated data to provide a high-quality annotated data set for subsequent model training; Specifically, a microalgae taxonomist can verify and optimize the annotated data to ensure the quality of the data set. The annotation tools are such as Labelbox, VGG Image Annotator (VIA), CVAT, etc.

[0020] S3: Based on the pre-trained weights of Grounded SAM, use the transfer learning method to fine-tune and optimize the model to adapt to microalgae detection; Grounded SAM is a deep learning model that combines object detection and image segmentation, integrating the capabilities of Grounding DINO (object detection) and Segment Anything Model (SAM, segmentation model). It can simultaneously achieve object localization (bounding box generation) and pixel-level segmentation. It supports out-of-the-box multi-object detection and can generate accurate segmentation masks, making it suitable for object detection and segmentation tasks in complex scenarios, such as microscopic image analysis and real-time monitoring.

[0021] Specifically, step S3 includes: S301: Format the labeled dataset in step S2 to meet the input requirements of the Grounded SAM model; S302: Select a widely trained Grounded SAM pre-trained model and adapt the trained model specifically to the microalgae detection task by fine-tuning the high-level features and output layer of the model; In deep learning, high-level features refer to the abstract and complex information extracted by the layers close to the output layer in the neural network.

[0022] The fine-tuning of the output layer includes: redefining the output of the classification layer and modulating the shape and size of the bounding box.

[0023] S303: During the training process, monitor the performance of the model through loss values and accuracy metrics, and adjust the model hyperparameters according to these metrics. The hyperparameters include learning rate, batch size, number of training epochs, and weights of the loss function to ensure the convergence and accuracy of model training; The accuracy metric can be mAP, i.e., mean average precision.

[0024] After training is completed, further optimize the model. The optimization methods used include pruning and quantization, etc., to improve the inference efficiency, save the weights of the fine-tuned model, and evaluate the model using the test set. The evaluation metrics include accuracy, recall, and F1 score.

[0025] S4: Preprocess the microscopic images collected in step S1, that is, after denoising using bilateral filtering, apply the Laplace operator to extract the edge information from the background and enhance the boundary features of the microalgae; Specifically, step S4 includes: S401: Convert the microscopic images collected in step S1, that is, RGB images into grayscale images ; S402: Apply bilateral filtering to the grayscale image Perform noise reduction while preserving the edge information of the image. The specific formula is as follows: where is the spatial distance weight, which is controlled by to control the smoothing range, is the pixel similarity weight, controls the similarity threshold, is the coordinate of the central pixel being processed currently; for a given point and a positive number ， the set of all points whose distance does not exceed is called the neighborhood of point ; represents the pixel coordinates within the neighborhood of the currently processed pixel , is the neighborhood window for the filtering operation. For example, in a window, and represent the position coordinates relative to the central pixel within the window, and , where represents the radius of the window.

[0026] S403: Perform a convolution operation on the denoised image using the Laplacian operator to extract edge information. The specific formula is as follows: , where is the Laplacian convolution kernel; where and represent the positions of the elements within, and the range of the index is determined by the size of . For example, for a 3×3 , and ∈{−1,0,1}.

[0027] S404: Superimpose the extracted edge information on the original image to generate an enhanced edge image , where is the edge enhancement weight coefficient.

[0028] S5: For the preprocessed microscopic image, based on the model trained in step S3, use Grounded SAM to implement object detection and segmentation of microalgae in the microscopic image, obtain the mask map corresponding to the microalgae, and segment the microalgae in the image through the mask map; Specifically, step S5 includes: S501: Use the image enhanced by the Laplacian operator As input, it is passed to the fine-tuned and optimized Grounded SAM model in step S3 for inference; S502: Use the Grounded SAM model to perform object detection on the input image, generate the bounding boxes of microalgae, extract the regions of interest, and output them as a set of class labels and a set of bounding boxes , where represent the upper left and lower right coordinates of each bounding box respectively; S503: Based on the feature extraction and classification of the target region, predict each pixel point in the enhanced image to generate the corresponding microalgae mask , where , where is a binary matrix; In the context of image processing and object detection, the "region of interest" (ROI) and the "target region" can overlap to some extent, but they are not exactly the same.

[0029] The target region is the area where the specific target identified by the object detection model (such as the Grounded SAM model) is located. That is to say, the target region refers to the area of the detected object or entity (microalgae) in the image.

[0030] The region of interest (ROI) is broader and refers to the area that the user or system is particularly concerned about. These areas are not necessarily the targets automatically detected by the object detection algorithm. They may be manually selected according to the task requirements or may be subsets of the target region. For example, the region of interest may be a part of the detected target region or an area that particularly needs further processing.

[0031] Participate Figure 5 to learn more about the binary mask map of the water area.

[0032] S504: Apply morphological operations to the segmentation mask of each target region to remove pseudo-targets and noise points, optimize the segmentation boundary, and the processed set of segmentation masks is: ; S505: Map each optimized target region mask back to the original coordinate system of the enhanced global image to generate the complete pixel-level global segmentation mask , where , and the global segmentation mask marks the pixel classes of all target regions in the microscopic image; S506: The enhanced original image Multiply pixel by pixel with the global segmentation mask to generate an image containing only the microalgae area , and .

[0033] S6: Count the types and volumes of microalgae, identify the types of algae, and calculate the algal density

[0034] Specifically, the step S6 includes: S601: Count the image containing only the microalgae area generated in step S506 and the target detection results, and accumulate the number of targets of each type of microalgae according to the pixel points belonging to the category in the segmentation mask: , where represents the target belongs to the category ; See Figure 6 for further understanding of the algal identification results, where the identification results are correct and the similarity is 0.64

[0035] S602: Estimate the volume of each target according to the pixel area of each segmentation mask and the magnification of the microscope: : where is the total number of mask pixel points, is the actual physical area corresponding to the unit pixel area determined by the microscope parameters, and summarize the volumes of all targets by category ; ; S603: Calculate the number of microalgae in the category according to the preset unit volume of microalgae in the experiment , and , and , the preset unit volume of microalgae is a fixed value determined by measuring standard samples; S604: Calculate the density of the category according to the sampling volume of the visible field of view of the microscope and the number of target types : : The total density of all targets is calculated as: .

[0036] As a further implementation method, the algal intelligent analysis method further includes: S7: Feed the calculated algae density result back to the user through a communication protocol.

[0037] As Figure 4 shown, an algae intelligent analysis system is used to implement the algae intelligent analysis method described in this application. The system includes a flow cytometer, an edge computing device, and a cloud server. The flow cytometer is connected to the edge computing device. The flow cytometer is configured with a flow cytometry microscopy device, which is used to capture high-resolution microscopic images of sub-visible and visible microalgae in the liquid flow of the microfluidic channel and upload the microscopic images to the edge computing device according to the sampling period. The edge computing device is used to perform object detection and segmentation processing on the microscopic images, that is, RGB images, at the edge. The edge computing device communicates with the cloud server through a network. The cloud server is used to identify algae and calculate the algae density.

[0038] The flow cytometer is used to collect microscopic images in the microfluidic channel based on the target water body. The edge computing device is used to deploy the trained Grounded SAM. Preprocess the labeled microscopic images, that is, after noise reduction using bilateral filtering, apply the Laplace operator to extract the edge information from the background to enhance the boundary features of the microalgae. For the preprocessed microscopic images, based on the Grounded SAM model, perform object detection and segmentation of the microalgae in the microscopic images, obtain the mask map corresponding to the microalgae, and segment the microalgae in the image through the mask map. The cloud server is used to count the types and volumes of microalgae, identify the algae species, calculate the algae density, and feedback the identification result and the algae density to the user through the network.

[0039] As a further implementation, the cloud server is also used to feedback the identification result and the algae density result to the user.

[0040] By adopting the Grounded DINO large model in this application, the accuracy and efficiency in image segmentation are greatly improved. At the same time, the image enhancement and other processing methods of the method in this application can be more conducive to the identification of segmented images.

[0041] In summary, this application provides an algae intelligent analysis method and system. By performing image segmentation and image enhancement processing on the images collected by the microscopic camera of the flow cytometer, the types of microalgae and the total volume of the corresponding microalgae are obtained. According to the volume of the microalgae in the annotation, the microalgae density information is further obtained, realizing the automation and intelligence of microalgae monitoring, improving the monitoring efficiency and accuracy, and at the same time realizing algal bloom early warning.

[0042] The above are only specific embodiments of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed in the present application can easily think of changes or substitutions, which should all be covered within the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.

Claims

1. An intelligent analysis method for algae, characterized in that: include: S1: Based on the target water body, the flow analyzer is used to collect microscopic images in the microfluidic channel; S2: Use annotation tools that support image segmentation and classification, and combine the corresponding information on microalgae classification to accurately annotate the category, location, and boundary features of microalgae in microscopic images, and verify and optimize the annotated data to provide a high-quality annotated data set for subsequent model training; S3: Based on the pre-trained weights of Grounded SAM, the model is fine-tuned and optimized using the transfer learning method to adapt to microalgae detection; S4: pre-processing the microscopic image collected in step S1, that is, using bilateral filtering to perform noise reduction, and then applying the Laplacian operator to extract edge information from the background to enhance the boundary features of the microalgae; S5: For the preprocessed microscopic image, Grounded SAM is used based on the model trained in step S3 to detect and segment the microalgae in the microscopic image, obtain the mask image corresponding to the microalgae, and segment the microalgae in the image through the mask image; S6: Count the types and volumes of microalgae, identify algae species, and calculate algae density.

2. The algae intelligent analysis method according to claim 1, characterized in that: The algae intelligent analysis method also includes: S7: Feedback the calculated algae density result to the user through the communication protocol.

3. The algae intelligent analysis method according to claim 2, characterized in that: The algae intelligent analysis method further includes: Step S3 specifically comprises: S301: Format the labeled data set in step S2 to meet the input requirements of the Grounded SAM model; S302: Select a Grounded SAM pre-trained model that has been extensively trained, and fine-tune the high-level features and output layers of the model to make the trained model specifically suitable for the microalgae detection task; S303: During the training process, the performance of the model is monitored through loss values ​​and accuracy indicators, and model hyperparameters are adjusted according to these indicators. The hyperparameters include learning rate, batch size, training cycle, and weight of loss function to ensure the convergence and accuracy of model training; S304: After training is completed, the model is further optimized using optimization methods including pruning and quantization to improve inference efficiency, the fine-tuned model weights are saved, and the model is evaluated using a test set, with evaluation indicators including accuracy, recall, and F1 value.

4. The algae intelligent analysis method according to claim 3, characterized in that: The step S4 is specifically as follows: S401: The microscopic image collected in step S1, i.e., the RGB image Convert to grayscale image ; S402: Use bilateral filtering to process grayscale images Perform noise reduction while retaining the edge information of the image. The specific formula is: in, is the spatial distance weight, given by Control the smoothing range, is the pixel similarity weight, Control the similarity threshold, is the pixel coordinate of the current processing center, are the coordinates of other pixels in the neighborhood; S403: Use the Laplace operator to perform a convolution operation on the denoised image to extract edge information. The specific formula is: ,in is the Laplace convolution kernel; S404: Superimposing the extracted edge information with the original image to generate an enhanced edge image ,in is the edge enhancement weight coefficient.

5. The algae intelligent analysis method according to claim 4, characterized in that: The step S5 is specifically as follows: S501: Using the image enhanced by the Laplacian operator As input, it is passed to the fine-tuned and optimized Grounded SAM model in step S3 for inference; S502: Use the Grounded SAM model to perform target detection on the input image, generate the bounding box of the microalgae, extract the region of interest, and output it as a set of category labels and bounding box collection ,in , representing the coordinates of the upper left corner and lower right corner of each bounding box; S503: Based on the feature extraction and classification of the target area, each pixel in the enhanced image is Make predictions and generate corresponding microalgae masks ,in, ,in is a binary matrix; S504: Segmentation mask for each target area Morphological operations are applied to remove pseudo targets and noise points and optimize segmentation boundaries. The processed segmentation mask set is: ; S505: Mask each optimized target area Mapping back to the original coordinate system of the enhanced global image generates a complete pixel-level global segmentation mask ,in , global segmentation mask Mark the pixel categories of all target areas in the microscopic image; S506: The enhanced original image With the global segmentation mask Multiply pixel by pixel to generate an image containing only the microalgae area ,and .

6. The algae intelligent analysis method according to claim 5, characterized in that: The step S6 is specifically as follows: S601: Counting the images containing only the microalgae region generated in step S506 and target detection results, the number of targets for each microalgae Accumulate the pixels belonging to the category in the segmentation mask: ,in Indicates the target Belongs to category ; S602: Based on each segmentation mask The volume of each object is estimated based on the pixel area and the magnification of the microscope : in is the total number of mask pixels, The actual physical area corresponding to the unit pixel area is determined by the microscope parameters. Sum up the volumes of all objects ; S603: According to the preset unit volume of microalgae in the experiment , calculation category The number of microalgae ,and , the preset microalgae unit volume It is a fixed value determined by a standard sample; S604: Sampling volume based on the visible field of view of the microscope and the number of target species , calculation category Density : The overall density of all targets is calculated as: .

7. An algae intelligent analysis system, characterized in that: The system is used to implement an intelligent algae analysis method as described in any one of claims 1 to 6, and the system includes a flow analyzer, an edge computing device and a cloud server; the flow analyzer is connected to the edge computing device; the flow analyzer is equipped with a flow microscopy imaging device, and the flow microscopy imaging device is used to capture high-resolution microscopic images of sub-visible and visible microalgae in the liquid flow of the microfluidic channel, and upload the microscopic images to the edge computing device according to the sampling period; the edge computing device is used to perform target detection and segmentation processing on the microscopic image, i.e., RGB image, at the edge; the edge computing device communicates with the cloud server through a network; and the cloud server is used to identify algae and calculate algae density.

Images