Chemical identification method based on convolutional neural network and large daphnia motion trajectory
By exposing large Daphnia magna to water and recording its behavioral trajectory, and using CNN to train a model, the problem of high cost and low accuracy in water chemical identification was solved, achieving efficient and low-cost pollutant identification.
Patent Information
- Application Number
- CN202211393275.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-11-08
- Publication Date
- 2025-11-04
- Estimated Expiration
- 2042-11-08
AI Technical Summary
Existing methods for identifying chemicals in water bodies are too costly and have low accuracy, making it difficult to achieve efficient and low-cost pollutant identification.
By exposing large daphnia to water using a variety of chemicals, recording their behavior using a trajectory recording system, converting the images into images and labeling them, and then training a model using a convolutional neural network (CNN) to optimize the network structure, the identification of chemical types and concentrations can be achieved.
It achieves high-accuracy, low-cost auxiliary identification of pollutants, reduces the cost of chemical identification in water bodies, and improves the accuracy of identification.
Smart Images
Figure CN115761579B_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of pollutant identification, and particularly relates to a chemical identification method based on a convolutional neural network and a large daphnia movement trajectory. BACKGROUND
[0002] Under the background of global social industrialization, water pollution has become one of the major environmental problems faced by many countries. Determining the chemical components in polluted water bodies is a basic work for the treatment of polluted water bodies. At present, chemical methods (such as chemical coloration) or instrumental analysis methods (such as mass spectrometry) are generally used for non-targeted identification of chemicals in water bodies, but these methods have the disadvantages of complicated procedures, low efficiency, poor targeting, and large chemical consumption. Therefore, it is urgent to develop new technologies to achieve convenient, efficient and green identification of water pollutants.
[0003] Compared with directly determining the structure of pollutants from a microscopic perspective, indirectly completing the auxiliary identification of water pollutants by extracting environmental information features seems to be a more simple and effective way. Studies have shown that aquatic organisms will exhibit unique movement trajectories when coexisting with pollutants. For example, lead acetate exposure causes daphnia magna to be in an "excited" state by affecting neural signal transmission and muscle fibers; fluoxetine antidepressants change the behavior of the anxiety-related behavior of the leopard fish by changing the serotonin system; pesticides interfere with the spatial memory of zebrafish and rare gobiocypris rarus, thereby affecting their behavior. Therefore, extracting the "trajectory fingerprint features" of organisms under different chemical exposures may serve as a basis for chemical identification. As one of the representative algorithms of deep learning, the convolutional neural network (CNN) can extract features from input data with a focus on convolutional computation, accurately and efficiently completing image recognition, target detection and other tasks. Therefore, training a CNN model to extract "fingerprint features" from biological movement trajectories is a feasible way.
[0004] The accuracy of the CNN training result often depends on the size of the data. However, there is currently no perfect biological movement trajectory database, which greatly limits the application of CNN models in biological movement trajectory feature extraction. Daphnia magna is a sentinel species widely distributed in freshwater lakes around the world and has been recognized internationally as a model organism in the field of ecological toxicology. As one of the most sensitive aquatic organisms to exogenous substances, daphnia magna has the advantages of being easy to observe with the naked eye and easy to cultivate in the laboratory, and has great advantages when used as a model organism for high-throughput exposure. Therefore, the model obtained by training the CNN using the biological movement trajectory dataset obtained through high-throughput daphnia magna live exposure is expected to complete the low-cost auxiliary identification of water chemicals.
[0005] Therefore, how to avoid the high cost and low accuracy of existing water body chemical identification is still a problem to be solved by those skilled in the art. SUMMARY
[0006] The application provides a chemical identification method based on a convolutional neural network and a daphnia magna motion trajectory, to solve the problems of high cost and low accuracy of existing water body chemical identification.
[0007] The application provides a chemical identification method based on a convolutional neural network and a daphnia magna motion trajectory, comprising:
[0008] A plurality of chemicals are selected for water exposure of daphnia magna;
[0009] After the exposure is completed, a trajectory recording system is used to record the behavior trajectory of daphnia magna;
[0010] The recorded trajectory video is converted into a picture, and a single daphnia magna trajectory picture within a fixed time is obtained by clipping;
[0011] The trajectory picture is labeled according to the chemical type and concentration to obtain a data set for training a chemical identification model;
[0012] A preset CNN network architecture is selected as the network structure of the chemical identification model to train the data set and optimize the parameters in the network structure, thereby obtaining a trained chemical identification model;
[0013] The trajectory picture of the daphnia magna to be identified is input into the trained chemical identification model, and the chemical type and concentration are output.
[0014] According to the chemical identification method based on a convolutional neural network and a daphnia magna motion trajectory provided by the application, the exposure environment conditions of the water-exposed daphnia magna are as follows: a salinity of 200 mg / L-230 mg / L, a pH value of 7.50±0.15, a temperature of 22±1℃, a light-dark cycle of 16h light plus 8h dark, and a light intensity of 2000lx.
[0015] According to the chemical identification method based on a convolutional neural network and a daphnia magna motion trajectory provided by the application, the recording of the behavior trajectory of daphnia magna by the trajectory recording system specifically comprises:
[0016] An animal behavior tracking system equipped with an infrared camera is used to record the behavior trajectory of daphnia magna.
[0017] According to the chemical identification method based on a convolutional neural network and a daphnia magna motion trajectory provided by the application, the recording of the behavior trajectory of daphnia magna by the trajectory recording system further comprises:
[0018] The behavior trajectory of Daphnia magna under dark condition and light condition was recorded for 15 min.
[0019] According to the chemical identification method based on the convolutional neural network and the motion trajectory of Daphnia magna provided by the application, the data set includes four categories, namely the motion trajectory graph of Daphnia magna under light condition within 60 s, the motion trajectory graph of Daphnia magna under light condition within 120 s, the motion trajectory graph of Daphnia magna under dark condition within 60 s, and the motion trajectory graph of Daphnia magna under dark condition within 120 s.
[0020] According to the chemical identification method based on the convolutional neural network and the motion trajectory of Daphnia magna provided by the application, the preset CNN network architecture is selected as the network structure of the chemical identification model to train the data set and optimize the parameters in the network structure, and specifically includes:
[0021] The residual network ResNet architecture is selected as the network structure of the chemical identification model to train the data set and optimize the parameters in the network structure, and the number of network layers in the ResNet architecture is 50.
[0022] According to the chemical identification method based on the convolutional neural network and the motion trajectory of Daphnia magna provided by the application, the training set of the chemical identification model during training: the data set parameter is set to 8:2, the batch size parameter is set to 24, and the step size parameter of the learning rate is set to 9.
[0023] The chemical identification method based on the convolutional neural network and the motion trajectory of Daphnia magna provided by the application, a plurality of chemicals are selected to expose Daphnia magna in water; after the exposure is completed, a trajectory recording system is used to record the behavior trajectory of Daphnia magna; the recorded trajectory video is converted into pictures, and a single Daphnia magna trajectory picture within a fixed time is obtained through clipping; the trajectory picture is labeled according to the chemical category and concentration to obtain a data set for training a chemical identification model; a preset CNN network architecture is selected as the network structure of the chemical identification model to train the data set and optimize the parameters in the network structure, and a trained chemical identification model is obtained; the trajectory picture of the Daphnia magna to be identified is input into the trained chemical identification model, and the chemical category and concentration are output. A high-accuracy low-cost auxiliary identification of pollutants is realized. BRIEF DESCRIPTION OF DRAWINGS
[0024] In order to make the technical solutions in the present application or the prior art clearer, the accompanying drawings needed in the embodiments or the prior art description will be briefly introduced. Obviously, the accompanying drawings in the following description are some embodiments of the present application, and other drawings can be obtained by those skilled in the art without any creative effort on the basis of these drawings.
[0025] Figure 1 A flowchart of the chemical identification method based on the convolutional neural network and the large Daphnia motion trajectory provided by the present application is shown in
[0026] Figure 2 A flowchart of the chemical identification technology based on the neural convolutional network and the large Daphnia motion trajectory provided by the present application is shown in
[0027] Figure 3 The accuracy rate graph of four data sets trained by different CNN models in 30 epochs provided by the present application is shown in
[0028] Figure 4 The accuracy rate graph of the optimized ResNet50 model provided by the present application is shown in
[0029] Figure 5 The Grad-CAM heat map of the ResNet50 and ResNet152 models provided by the present application is shown in DETAILED DESCRIPTION
[0030] In order to make the technical solutions in the present application or the prior art clearer, the accompanying drawings needed in the embodiments or the prior art description will be briefly introduced. Obviously, the accompanying drawings in the following description are some embodiments of the present application, and other drawings can be obtained by those skilled in the art without any creative effort on the basis of these drawings.
[0031] Since the existing water body chemical identification generally has the problems of high cost and low accuracy. The chemical identification method based on the convolutional neural network and the large Daphnia motion trajectory provided by the present application will be described below. Figure 1 The chemical identification method based on the convolutional neural network and the large Daphnia motion trajectory provided by the present application is described. Figure 1 A flowchart of the chemical identification method based on the convolutional neural network and the large Daphnia motion trajectory provided by the present application is shown in Figure 1 As shown in the figure, the method comprises:
[0032] Step 110, a plurality of chemicals are selected to expose the large Daphnia to the water body.
[0033] Specifically, 32 chemicals are selected in the embodiment of the present application, including 6 heavy metals (nickel chloride, methyl mercury chloride, mercury chloride, copper sulfate, cadmium chloride, lead acetate), 16 flame retardants (tris (1,3-dichloro-2-propyl) phosphate, tris (2-chloroethyl) phosphate, tris (2-chloropropyl) phosphate, triphenyl phosphate, perfluorooctanoic acid, perfluorooctane sulfonic acid, 2,2',4,4'-tetrabromodiphenyl ether, decabromodiphenyl ether, hexabromocyclododecane, 4,4'-dihydroxydiphenyl ether, tetrabromobisphenol A, tetrachlorobisphenol A, bisphenol A, bisphenol S, bisphenol F, bisphenol AF), 6 pesticides (abamectin, thiamethoxam, imidacloprid, chlorpyrifos, lambda-cyhalothrin, methyl parathion) and 4 fungicides (prochloraz, 1,2-benzisothiazolin-3-one, dazomet, clotrimazole). These chemicals are often detected in aquatic environments due to their high production and wide use in various fields. Daphnia magna is cultured in filtered pure water. The water quality indicators and feeding environment are as follows: salinity (200 mg / L-230 mg / L), pH (7.50±0.15), temperature (22±1℃), light-dark cycle (16h light / 8h dark), light intensity (2000lx). After sexual maturity, more than three offspring are required for the experiment. Juvenile Daphnia magna is used for exposure experiments. The 48h 5% effective concentration (EC5), 1 / 4EC5 and 1 / 16EC5 of each chemical are selected as the exposure concentration. A 48-well plate is used as the chemical exposure container. One 48-well plate is used for each concentration, and each well contains a single Daphnia magna. The Daphnia magna in the plate is exposed for 48h.
[0034] Step 120, after the exposure is completed, the trajectory recording system is used to record the behavior trajectory of Daphnia magna.
[0035] Specifically, after the exposure is completed, Daphnia magna is transferred to an observation room, and an animal behavior tracking system is used to record the behavior trajectory of Daphnia magna. The system is equipped with an infrared camera, which facilitates tracking Daphnia magna in dark conditions. The dark condition and the light condition are each recorded for 15 minutes, for a total of 30 minutes.
[0036] Step 130, the recorded trajectory video is converted into pictures, and the trajectory pictures of a single Daphnia magna within a fixed time are obtained by clipping.
[0037] Specifically, after the Daphnia magna behavior video is recorded, the EthoVision XT behavior analysis software is used to extract the trajectory, and after the screenshot of the motion trajectory is obtained, the Python program is used to extract the pure trajectory graph and clip the motion trajectory of each Daphnia magna, forming a large number of 125*125 resolution pictures.
[0038] Step 140, the trajectory pictures are labeled according to chemical categories and concentrations to obtain a data set for training a chemical identification model.
[0039] Specifically, the trajectory pictures are labeled according to chemical categories and concentrations to obtain a data set for training a chemical identification model.
[0040] Step 150, a preset CNN network architecture is selected as the network structure of the chemical identification model to train the data set and optimize parameters in the network structure, thereby obtaining a trained chemical identification model.
[0041] Specifically, an appropriate network architecture is selected as the network structure of the chemical identification model for training, and the to-be-adjusted parameters in the network structure are adjusted to optimal values, thereby obtaining a trained chemical identification model.
[0042] Step 160, inputting the trajectory pictures of the daphnia magna to be identified into the trained chemical identification model to output chemical categories and concentrations.
[0043] Specifically, the trained chemical identification model is put into use, that is, the trajectory pictures of the daphnia magna to be identified are input into the trained chemical identification model, and the model outputs the chemical categories and concentrations corresponding to the pictures.
[0044] The chemical identification method based on the convolutional neural network and the daphnia magna movement trajectory provided by the application comprises the following steps: exposing daphnia magna to water by selecting multiple chemicals; recording the behavior trajectory of the daphnia magna by using a trajectory recording system after the exposure ends; converting the recorded trajectory video into pictures, and obtaining the trajectory pictures of a single daphnia magna in a fixed time by clipping; labeling the trajectory pictures according to chemical categories and concentrations to obtain a data set for training a chemical identification model; selecting a preset CNN network architecture as the network structure of the chemical identification model to train the data set and optimize parameters in the network structure, thereby obtaining a trained chemical identification model; and inputting the trajectory pictures of the daphnia magna to be identified into the trained chemical identification model to output chemical categories and concentrations. The method realizes the low-cost auxiliary identification of pollutants with high accuracy.
[0045] Based on the above embodiment, the exposure environment conditions of the daphnia magna exposed to water are as follows: a salinity of 200 mg / L-230 mg / L, a pH value of 7.50±0.15, a temperature of 22±1℃, a light-dark cycle of 16h light plus 8h dark, and a light intensity of 2000lx.
[0046] Specifically, the daphnia magna needs to be bred, and the water environment for breeding is a salinity of 200 mg / L-230 mg / L, a pH value of 7.50±0.15, a temperature of 22±1℃, a light-dark cycle of 16h light plus 8h dark, and an illumination intensity of 2000lx. After sexual maturity, more than three offspring are needed to be used for experiments. The daphnia magna larvae are used for exposure experiments.
[0047] Based on any one of the above embodiments, in the method, the recording of the behavior trajectory of the daphnia magna by the trajectory recording system specifically includes:
[0048] The behavior trajectory of the daphnia magna is recorded by using an animal behavior tracking system equipped with an infrared camera.
[0049] Specifically, after the exposure ends, the daphnia magna is transferred to an observation room, and the behavior trajectory of the daphnia magna is recorded by using an animal behavior tracking system. The system is equipped with an infrared camera.
[0050] Based on any one of the above embodiments, in the method, the recording of the behavior trajectory of the daphnia magna by the trajectory recording system further includes:
[0051] The behavior trajectory of the daphnia magna under dark conditions and light conditions is recorded for 15 minutes respectively.
[0052] Specifically, the animal behavior tracking system is equipped with an infrared camera, which facilitates tracking of the daphnia magna under dark conditions. The dark conditions and the light conditions are recorded for 15 minutes respectively, for a total of 30 minutes.
[0053] Based on any one of the above embodiments, in the method, the data set includes four categories, which are a motion trajectory graph of the daphnia magna under light conditions within 60s, a motion trajectory graph of the daphnia magna under light conditions within 120s, a motion trajectory graph of the daphnia magna under dark conditions within 60s, and a motion trajectory graph of the daphnia magna under dark conditions within 120s.
[0054] Specifically, after the behavior video recording of Daphnia magna is completed, the trajectory is extracted using the EthoVision XT behavior analysis software. In order to explore the optimal training condition, four data sets are collected, which are: the movement trajectory graph of Daphnia magna for 60s in the dark condition (Dark60s); the movement trajectory graph of Daphnia magna for 120s in the dark condition (Dark120s); the movement trajectory graph of Daphnia magna for 60s under light condition (Light60s); and the movement trajectory graph of Daphnia magna for 120s under light condition (Light120s). After the screenshot of the movement trajectory is obtained, the pure trajectory graph is extracted using a Python program, and the movement trajectory of each Daphnia magna is cropped to form a large number of 125*125 resolution pictures. In the exposure experiment, if Daphnia magna is not moving, a blank picture will be obtained at last, so the images with a file size less than 1.5KB are deleted. The number of images in the four data sets (Light60s, Dark60s, Light120s and Dark120s) used for CNN training is 160509, 144953, 128801 and 121983 respectively.
[0055] Based on any one of the above embodiments, in the method, the preset CNN network architecture is selected as the network structure of the chemical identification model to train the data set and optimize parameters in the network structure, and specifically includes:
[0056] The residual network ResNet architecture is selected as the network structure of the chemical identification model to train the data set and optimize parameters in the network structure, and the number of network layers in the ResNet architecture is 50.
[0057] Specifically, a plurality of network architectures are selected as the training network of the chemical identification model, and the network architecture with the highest accuracy is selected through experiments, which is the residual network ResNet architecture with 50 network layers, i.e. the ResNet architecture with 50 network layers is used as the network structure of the chemical identification model to train the data set and optimize parameters in the network structure.
[0058] Based on any one of the above embodiments, in the method, the training set when the chemical identification model is trained: the data set parameter is set to 8:2, the batch size parameter is set to 24, and the step size parameter of the learning rate is set to 9.
[0059] Specifically, it is found through experiments that ResNet50 has the highest accuracy on the Light120s dataset, which can reach 83.9%. Using the optimized parameters, that is, the training set: dataset is 8:2, batch size is 24, and the learning rate adjustment interval step size is 9, ResNet50 can obtain an accuracy of 88.3% on the Light120s dataset.
[0060] Based on the above embodiments, the application provides a chemical identification technology based on a neural convolutional network and a large Daphnia motion trajectory, Figure 2 The flowchart of the chemical identification technology based on a neural convolutional network and a large Daphnia motion trajectory provided by the application is shown in Figure 2 The technology comprises:
[0061] (1) Chemical exposure
[0062] In this case, 32 kinds of chemicals are selected, including 6 heavy metals (nickel chloride, methyl mercuric chloride, mercuric chloride, copper sulfate, cadmium chloride, and lead acetate), 16 flame retardants (tris (1,3-dichloro-2-propyl) phosphate, tris (2-chloroethyl) phosphate, tris (2-chloropropyl) phosphate, triphenyl phosphate, perfluorooctanoic acid, perfluorooctane sulfonic acid, 2,2',4,4'-tetrabromodiphenyl ether, decabromodiphenyl ether, hexabromocyclododecane, 4,4'-dihydroxydiphenyl ether, tetrabromobisphenol A, tetrachlorobisphenol A, bisphenol A, bisphenol S, bisphenol F, and bisphenol AF), 6 pesticides (abamectin, thiamethoxam, imidacloprid, chlorpyrifos, lambda-cyhalothrin, and methyl parathion), and 4 fungicides (prochloraz, 1,2-benzisothiazol-3-ketone, dazomet, and clotrimazole). These chemicals are often detected in aquatic environments due to their high production and widespread use in various fields. The Daphnia magna is cultured in filtered pure water. The water quality indicators and feeding environment are as follows: salinity (200 mg / L-230 mg / L), pH (7.50±0.15), temperature (22±1℃), light and dark cycle (16h light / 8h dark), and light intensity (2000lx). After sexual maturity, the offspring can be used for experiments if they have more than three offspring. The Daphnia magna larvae are used for exposure experiments. The 48h 5% effective concentration (EC5), 1 / 4EC5, and 1 / 16EC5 of each chemical are selected as the exposure concentrations. A 48-well plate is used as the chemical exposure container. One 48-well plate is used for each concentration, and each well contains a single Daphnia magna. The Daphnia magna in the plate is exposed for 48h.
[0063] (2) Behavior trajectory recording
[0064] After the exposure, the daphnia magna were transferred to the observation chamber, and the animal behavior tracking system was used to record the behavior trajectory of the daphnia magna. The system is equipped with an infrared camera, which facilitates tracking of daphnia magna in dark conditions. Dark conditions and light conditions were recorded for 15 minutes each, for a total of 30 minutes.
[0065] (3) Data set collection
[0066] After the behavior video of daphnia magna was recorded, the EthoVision XT behavior analysis software was used to extract the trajectory. In order to explore the optimal training conditions, four data sets were collected, namely: the movement trajectory graph of daphnia magna for 60s in dark conditions (Dark60s); the movement trajectory graph of daphnia magna for 120s in dark conditions (Dark120s); the movement trajectory graph of daphnia magna for 60s in light conditions (Light60s); the movement trajectory graph of daphnia magna for 120s in light conditions (Light120s). After obtaining the screenshot of the movement trajectory, a Python program was used to extract the pure trajectory graph, and the movement trajectory of each daphnia magna was cropped to form a large number of 125*125 resolution images. In the exposure experiment, if the daphnia magna is not moving, a blank image will be obtained at the end, so images with file size less than 1.5KB are deleted. The number of images in the four data sets (Light60s, Dark60s, Light120s and Dark120s) used for CNN training is 160509, 144953, 128801 and 121983 respectively.
[0067] (4) CNN model training
[0068] In this case, ResNet (ResNet18, 34, 50, 101 and 152) and DenseNet (DenseNet121, 161, 169 and 201) architectures are used for CNN training. Different numbers represent different network layers. The deep learning framework uses the Python-based deep learning library PyTorch, the GPU model is NVIDIA 3070Ti with 8G of video memory; the input image size is 125*125; the learning rate is adjusted at equal intervals, the adjustment multiple gamma is set to 0.1, and the adjustment interval step size is set to 7; the optimizer uses Ranger; the batch size is set to 16; the epoch is set to 30. The accuracy is used as the evaluation index of the classification result.
[0069] Table 1 shows the best accuracy of different CNN models provided by the present application in 30 epochs, Figure 3 Figure 1 shows the accuracy of four data sets in 30 epochs for different CNN models provided by the present application, Figure 4The accuracy graph of the optimized ResNet50 model provided by the present application, Figure 5 The Grad-CAM heat map of the ResNet50 and ResNet152 models provided by the present application.
[0070] Table 1: Best accuracy of different CNN models trained for 30 epochs
[0071]
[0072] The best accuracy of different CNN models trained on four datasets is shown in Table 1 and Figure 3 The results show that ResNet50 has the highest accuracy on the Light120s dataset, reaching 83.9%. Using the optimized parameters, i.e., training set: dataset ratio of 8:2, batch size of 24, and learning rate adjustment interval step size of 9, ResNet50 can achieve an accuracy of 88.3% on the Light120s dataset, as shown in Figure 4 The accuracy of the ResNet50 model is shown in Figure 5 The Grad-CAM heat map shows that ResNet50 effectively extracts the motion trajectory features of Daphnia magna, indicating the reliability of the model in the auxiliary identification of water pollutants.
[0073] Finally, it should be noted that the above examples are only used to illustrate the technical solutions of the present application, and are not limited thereto; although the present application has been described in detail with reference to the foregoing examples, those skilled in the art should understand that they can still modify the technical solutions described in the foregoing examples, or make equivalent substitutions for some technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present application.
Claims
1. A chemical identification method based on convolutional neural network and large Daphnia motion trajectory, characterized in that, The method comprises the following steps: selecting a plurality of chemicals to expose daphnia magna in water; recording the behavior trajectory of daphnia magna using a trajectory recording system after the exposure ends; converting the recorded trajectory video into pictures, and obtaining single daphnia magna trajectory pictures in a fixed time by clipping; labeling the trajectory pictures according to the type and concentration of the chemicals to obtain a data set for training a chemical recognition model; selecting a preset CNN network architecture as the network structure of the chemical recognition model to train the data set and optimize the parameters in the network structure, thereby obtaining a trained chemical recognition model; inputting the trajectory pictures of daphnia magna to be identified into the trained chemical recognition model to output the type and concentration of the chemicals.
2. The chemical identification method based on convolutional neural network and large Daphnia motion trajectory according to claim 1, characterized in that, The exposure environment conditions of the daphnia magna exposed in water are as follows: salinity of 200 mg / L-230 mg / L, pH value of 7.50±0.15, temperature of 22±1℃, light-dark cycle of 16h light plus 8h dark, and light intensity of 2000lx. 3.The method according to claim 1, wherein, The recording of the behavior trajectory of daphnia magna using a trajectory recording system specifically comprises: recording the behavior trajectory of daphnia magna using an animal behavior tracking system equipped with an infrared camera.
4. The method according to claim 1, wherein, The recording of the behavior trajectory of daphnia magna using a trajectory recording system further comprises: recording the behavior trajectory of daphnia magna under dark conditions and light conditions for 15 minutes respectively. 5.The method according to claim 1, wherein, The data set comprises four categories, namely the movement trajectory graph of daphnia magna within 60s under light conditions, the movement trajectory graph of daphnia magna within 120s under light conditions, the movement trajectory graph of daphnia magna within 60s under dark conditions, and the movement trajectory graph of daphnia magna within 120s under dark conditions. 6.The method according to claim 1, wherein, The selection of a preset CNN network architecture as the network structure of the chemical recognition model to train the data set and optimize the parameters in the network structure specifically comprises: selecting a residual network ResNet architecture as the network structure of the chemical recognition model to train the data set and optimize the parameters in the network structure, wherein the number of network layers in the ResNet architecture is 50.
7. The method according to claim 6, wherein, The training set of the chemical recognition model during training: the data set parameter is set to 8:2, the batch size parameter is set to 24, and the step size parameter of the learning rate is set to 9.