Sample automatic processing system and method based on neural network

Abstract

The present invention discloses a sample automatic processing system and method based on a neural network, which belongs to the field of artificial intelligence and machine learning technology, and specifically includes: capturing sample images through a scanner, applying convolutional neural networks and semantic segmentation technology to locate and separate DNA sample areas; combining an automated workstation and real-time fluorescence quantitative method, the system extracts DNA from samples, and monitors and records extraction process parameters in real time. The extracted DNA is subjected to spectral analysis, and the obtained data is fused with the sample image and process parameters, and input into a pre-trained classification neural network model for training optimization; the trained model is embedded in a control system, and combined with reinforcement learning to realize automatic detection of DNA extraction quality and parameter adjustment; efficient and intelligent processing of forensic evidence samples is achieved, and the accuracy and efficiency of DNA analysis are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of artificial intelligence and machine learning, and specifically to a sample automatic processing system and method based on a neural network. Background Art

[0002] As the requirements for accuracy and efficiency of sample analysis in the fields of biomedicine, forensic medicine, materials science and industrial production continue to increase, traditional manual sample processing methods can no longer meet the needs of large-scale, high-throughput analysis. Manual processing is not only time-consuming and labor-intensive, but also prone to human errors, affecting the reliability and repeatability of experimental results. The rapid development of neural networks, especially deep learning technology, has provided a new solution to this challenge. Neural networks have powerful pattern recognition and data analysis capabilities, and can automatically learn features from complex, high-dimensional data to achieve accurate identification, classification and even prediction of samples. At the same time, combined with machine vision, robotic operation and automated equipment, the system can automatically execute the entire process from sample reception, preprocessing, detection to data analysis, significantly improving processing speed and accuracy.

[0003] However, most existing automated sample processing systems rely on preset rules and fixed algorithms, lack flexibility and adaptability, and have difficulty coping with the diversity and complex changes of sample types. In addition, traditional methods often face problems of low recognition accuracy and low processing efficiency when dealing with samples with strong heterogeneity and various morphologies.

[0004] For example, the patent with publication number CN114036296A discloses a method for generating adversarial samples of black-box recurrent neural networks, including: extracting weighted finite automata, initializing a candidate sample set, predicting data, calculating fitness values, judging target thresholds, and generating adversarial samples. Based on this technical solution, adversarial data generation of natural language text is performed, and weighted finite automata are extracted from the target black-box recurrent neural network as a guide for adversarial sample generation of natural language text, providing proxy model guidance, increasing the success rate and efficiency of adversarial sample generation, and using this technical solution for black-box recurrent neural network adversarial sample generation in machine learning as a service scenarios, increasing the success rate and efficiency of adversarial sample generation, effectively reducing the number of inquiries, and reducing service inquiry costs.

[0005] For example, a Chinese patent with authorization announcement number CN109522454B discloses a method for automatically generating web sample data, including: constructing a sample training set of web training samples, performing feature processing on the web training samples in the sample training set to obtain corresponding encoding strings, and constructing a neural network model; inputting the encoding strings corresponding to the sample training set into the neural network model, calculating the loss function value, and optimizing the neural network model by minimizing the loss function value to obtain the optimal model; constructing a sample input set of web input samples, performing feature processing on the web input samples in the sample input set to obtain corresponding encoding strings, and inputting the encoding strings corresponding to the sample input set into the optimal model to generate a digital character sequence; converting the digital character sequence into a corresponding vocabulary character sequence. This technical solution is suitable for generating web sample data in security testing.

[0006] The above existing technologies all have the following problems: 1) They rely on DFA extracted from the black box model. Therefore, for the black box model where DFA cannot be effectively extracted, or when DFA does not completely correspond to the model behavior, this method cannot work effectively; 2) The generated adversarial samples are only effective for specific models or specific input conditions, and cannot maintain adversarial properties for other models or conditions; 3) The generated sample data cannot fully reflect this real-time and dynamic nature; 4) There are domain limitations and security vulnerabilities. Summary of the invention

[0007] In view of the shortcomings of the prior art, the present invention proposes a neural network-based automatic sample processing system and method, which captures sample images through a scanner, and uses convolutional neural networks and semantic segmentation technology to locate and separate DNA sample areas; combined with an automated workstation and real-time fluorescence quantification, the system extracts DNA from samples and monitors and records the extraction process parameters in real time. The extracted DNA is subjected to spectral analysis, and the resulting data is fused with the sample image and process parameters, and input into a pre-trained classification neural network model for training and optimization; the trained model is embedded in the control system, and combined with reinforcement learning to achieve automatic detection of DNA extraction quality and parameter adjustment; efficient and intelligent processing of forensic evidence samples is achieved, and the accuracy and efficiency of DNA analysis are improved.

[0008] To achieve the above object, the present invention provides the following technical solutions:

[0009] The automatic sample processing method based on neural network includes:

[0010] Step S1: Capturing a forensic evidence sample image by a scanner, extracting physical property data, contamination degree information, evidence preservation condition information, parameter annotation data, and DNA spectrum data of skin cells in the forensic evidence sample according to the forensic evidence sample image, and performing multimodal fusion to generate evidence sample data;

[0011] Step S2: input the physical evidence sample data into the pre-trained classification neural network model for training, and optimize the classification neural network model parameters using the incremental learning strategy to obtain the optimized classification neural network model;

[0012] Step S3: embed the optimized classification neural network model into the control system of the automated workstation, and train a policy network in combination with reinforcement learning. After the DNA extraction is completed, the extracted DNA is quality tested using PCR amplification and electrophoresis methods;

[0013] Step S4: If the DNA quality meets the standard, the result is fed back to the classification neural network model for further training and optimization. If the DNA quality does not meet the standard, the automated workstation automatically adjusts the extraction parameters or re-extracts according to the strategy network. At the same time, the data in the database and the classification neural network model parameters are regularly updated.

[0014] The specific steps of step S2 include:

[0015] S2.1: Obtaining physical evidence sample data using feature concatenation ,in, Represents the physical property data of skin cells, Indicates the degree of skin cell pollution. Displays the storage condition information of skin cells. Represents the parameter annotation data of skin cells, represents the DNA spectrum data of skin cells, a, b, c, d, and e represent the physical property data of skin cells, the pollution degree information, the storage condition information, the parameter annotation data, and the quantity of DNA spectrum data, respectively;

[0016] S2.2: Select a pre-trained classification neural network model, take the physical evidence sample data and corresponding labels as input, and use an incremental learning strategy to optimize and update the parameters of the classification neural network model.

[0017] S2.3: Evaluate the performance of the classification neural network model by calculating the accuracy, and adjust the structure, parameters or data preprocessing method of the classification neural network model according to the evaluation results;

[0018] The specific formula of the incremental learning strategy in S2.2 is:

[0019]

[0020] in, and represents the updated weights and biases of the lth layer, and represents the weight and bias of the lth layer, represents the error term of the lth layer, represents the modulation factor of the lth layer, represents the weight matrix, represents the weighted output of the lth layer, represents the derivative of the activation function, represents the learning rate, , They represent the momentum hyperparameters at the tth iteration in the Adam optimizer, represents the updated momentum term, represents the coefficient of the L2 regularization term, It represents the error term of the ith data in the physical evidence sample data X at the lth level, represents the Hadamard product, t represents the current iteration number, and m represents the number of data in the physical evidence sample data;

[0021] The specific formula of the policy network in step S3 is:

[0022]

[0023] in, represents the policy gradient, represents the parameter set of the nth policy network, n represents the number of policy network parameter sets, represents the state distribution under the strategy, represents the original advantage function, h represents the action, s represents the state, represents the policy network, represents the distribution of context information k under a given state s, represents the weight of the jth parameter set, represents the hyperparameter, represents the reward function, represents the expected function, represents the gradient, Represents a combination set;

[0024] The specific formula for quality detection in step S3 is:

[0025]

[0026] in, Indicates DNA quality, Indicates the number of DNA copies at the beginning of the PCR reaction. represents the molecular weight of a single DNA copy, q represents the amplification efficiency, and u represents the number of cycles in the PCR reaction. represents the recovery efficiency of PCR products in subsequent processing, and , Represents the purity factor.

[0027] Specifically, the physical property data of the skin cells is obtained by performing target detection on the forensic evidence sample image based on an improved convolutional neural network, locating the position of the forensic evidence sample in the image, and obtaining it according to the target detection and positioning information; the contamination degree information is obtained by separating the DNA sample area through semantic segmentation, identifying the pollutant area, and evaluating the preservation status of the skin cells in the pollutant area, and then obtaining the contamination degree of the skin cells according to the evaluation results; the evidence preservation condition information is obtained according to the environment in which the skin cells are located; the parameter annotation data is obtained by using an automated workstation combined with a real-time fluorescence quantification method to extract DNA from skin cells, monitoring the DNA extraction process, automatically recording and annotating the process parameters of the DNA extraction, and forming parameter annotation data; the DNA spectral data is obtained by performing spectral analysis on the extracted DNA to obtain DNA spectral data.

[0028] The sample automatic processing system based on neural network includes: image processing module, control module, evaluation module and interaction module;

[0029] The image processing module is used to preprocess, analyze and extract features from the image data of the forensic evidence sample;

[0030] The control module is used to automatically adjust the parameters of the processing equipment according to the outputs of the image processing module and the evaluation module, and perform the actual operation of physical evidence sample processing;

[0031] The evaluation module is used to perform quality inspection on the processed physical evidence samples, perform performance evaluation on the classification neural network model, and feed back the quality inspection results and performance evaluation results to the control module and the image processing module to adjust the control strategy or model parameters;

[0032] The interaction module is used to provide an interface between the user and the system, allowing the user to input instructions, view processing progress, and receive system feedback.

[0033] Specifically, the control module includes: an adaptive control unit, a decision-making unit, and an optimization unit;

[0034] The adaptive control unit is used to dynamically adjust the parameters of the processing device according to the output results of the classification neural network model;

[0035] The decision-making unit is used to formulate the next operation instruction based on the model prediction and real-time monitoring data;

[0036] The optimization unit is used to collect processing results and optimize the control strategy and model parameters through an optimization algorithm.

[0037] An electronic device includes a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of a neural network-based automatic sample processing method when executing the computer program.

[0038] A computer-readable storage medium stores computer instructions, which, when executed, execute the steps of a neural network-based automatic sample processing method.

[0039] Compared with the prior art, the present invention has the following beneficial effects:

[0040] 1. The present invention proposes an automatic sample processing system based on a neural network, and optimizes and improves the architecture, operating steps and processes. The system has the advantages of simple processes, low investment and operating costs, and low production work costs.

[0041] 2. The present invention proposes an automatic sample processing method based on a neural network. By using the forensic evidence sample images captured by a scanner and combining them with multimodal fusion technology, the feature information in the evidence samples can be more comprehensively extracted and integrated, thereby improving the accuracy and efficiency of data processing. The parameters of the classification neural network model are optimized through an incremental learning strategy, so that the model can continuously learn and adapt to new data, thereby improving the classification performance and generalization ability of the model. In addition, the optimized model is embedded in the control system of the automated workstation, further realizing the automation and intelligence of evidence processing.

[0042] 3. The present invention proposes an automatic sample processing method based on a neural network, combined with a strategy network trained by reinforcement learning, which automatically performs quality inspection after the DNA extraction is completed, and automatically adjusts the extraction parameters or re-extracts according to the test results, which not only improves the efficiency and success rate of DNA extraction, but also reduces human intervention and errors; regularly updates the data in the database and the classification neural network model parameters to ensure that the system always uses the latest data and models to work, thereby maintaining the advancement and accuracy of the system and improving the intelligence level of forensic evidence processing. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 Schematic diagram of the automatic sample processing method based on neural network of the present invention;

[0044] Figure 2 This is a workflow diagram of the automatic sample processing method based on neural network of the present invention;

[0045] Figure 3 This is a flow chart of DNA quality detection of the automatic sample processing method based on neural network of the present invention;

[0046] Figure 4 This is an architecture diagram of the sample automatic processing system based on neural network of the present invention. DETAILED DESCRIPTION

[0047] Example 1

[0048] See also Figure 1-Figure 3 , an embodiment of the present invention provides: a method for automatically processing samples based on a neural network, comprising the following steps:

[0049] Step S1: Capture the forensic evidence sample image by means of a scanner, perform target detection on the forensic evidence sample image by means of an improved convolutional neural network, locate the position of the forensic evidence sample in the image, and identify the physical characteristics and contamination degree of the skin cells, separate the DNA sample area by means of semantic segmentation, identify the pollutant area, and evaluate the preservation state of the skin cells, and at the same time, obtain the evidence preservation condition information according to the environment, and at the same time, use an automated workstation in combination with a real-time fluorescence quantitative method to extract DNA from the skin cells, monitor the DNA extraction process, automatically record and annotate the process parameters of the DNA extraction, form parameter annotation data, perform spectral analysis on the extracted DNA, obtain DNA spectral data, and perform multimodal fusion of the physical characteristics of the skin cells, the contamination degree, the evidence preservation condition information, the parameter annotation data, and the DNA spectral data to generate evidence sample data;

[0050] Among them, the process parameters of DNA extraction include: extraction conditions and abnormal phenomena.

[0051] Furthermore, the specific steps of step S1 include:

[0052] (1) Use a scanner to capture forensic evidence sample images, and use the improved convolutional neural network to perform target detection on the forensic evidence sample images:

[0053] 1) Design a CNN structure suitable for forensic evidence sample image detection, including multiple convolutional layers, pooling layers, and fully connected layers;

[0054] 2) Use a labeled forensic evidence sample image dataset to train the CNN model so that the model can learn and identify key targets in the image, such as skin cells and DNA samples;

[0055] 3) Input the captured forensic evidence sample image into the trained CNN model, and the model will output the location information of the target in the image.

[0056] (2) Based on the target detection results, identify the physical characteristics and contamination level of skin cells and evaluate the preservation status of skin cells;

[0057] 1) Extract features related to the physical properties of skin cells and the degree of contamination from the images using CNN or other image processing techniques;

[0058] 2) Based on the extracted features, the preservation status of skin cells was classified and evaluated using a support vector machine classification algorithm.

[0059] (3) Based on the classification results, the semantic segmentation method is used to separate the DNA sample area and identify the contaminant area;

[0060] 1) Design a semantic segmentation network suitable for forensic evidence sample images, and U-Net is selected in the present invention;

[0061] 2) Use a dataset with pixel-level labels to train a semantic segmentation network so that the network can learn and identify different areas in the image, such as DNA sample areas and pollutant areas. Then input the captured image into the trained semantic segmentation network, which will output the category label for each pixel, thereby separating and identifying the DNA sample area and the pollutant area.

[0062] (4) Obtaining information on the conditions for preserving physical evidence based on the environment, including collecting environmental information through temperature, humidity, and light sensors for real-time monitoring and recording of information on the conditions for preserving physical evidence.

[0063] The specific steps of step S1 also include:

[0064] (5) Sample preparation: Collect skin cell samples and perform pretreatment, including cleaning and crushing;

[0065] (6) Automated DNA extraction:

[0066] 1) Place the processed skin cell sample into the automated workstation;

[0067] 2) The automated workstation automatically adds the required reagents and solutions, such as lysis buffer and proteinase K, according to preset programs and parameters;

[0068] 3) The automated workstation controls temperature and time conditions to extract and purify DNA.

[0069] (7) Real-time fluorescence quantitative monitoring: During the DNA extraction process, the real-time fluorescence quantitative system monitors the concentration and purity of DNA by adding fluorescent probes or dyes, automatically records the fluorescence signal value at each time point, and forms the process parameter data of DNA extraction;

[0070] (8) Parameter annotation: The data of the fluorescence signal value recorded by the automated workstation is associated with the steps and conditions in the extraction process to form parameter annotation data;

[0071] (9) Spectral analysis: The extracted DNA samples are sent to a spectral analysis instrument for analysis. By measuring the absorbance or emission spectrum of the DNA samples at different wavelengths, the spectral data of the DNA is obtained, which is used to evaluate the purity and structural information of the DNA.

[0072] Step S2: input the physical evidence sample data into the pre-trained classification neural network model for training, optimize the classification neural network model parameters using the back propagation algorithm and incremental learning strategy, and obtain the optimized classification neural network model;

[0073] Step S3: embed the optimized classification neural network model into the control system of the automated workstation, and train a policy network in combination with reinforcement learning. After the DNA extraction is completed, the extracted DNA is quality tested using PCR amplification and electrophoresis methods;

[0074] Further, the PCR amplification step comprises:

[0075] (1) Prepare the reaction system: Add the components, including DNA template, specific primers, dNTP Mix, PCR buffer, and hot-start enzyme, into a 0.2 mL centrifuge tube in sequence. If a hot-start enzyme is used, the reaction can be performed at room temperature. Non-hot-start enzymes need to be prepared at low temperature to prepare a 25 μL reaction system.

[0076] (2) Set the PCR instrument program: set the reaction time and temperature according to the kit instructions, including pre-denaturation at 95°C, denaturation at 95°C, annealing at 56°C, extension at 72°C, and complementary extension at 72°C, and cycle 30 times;

[0077] (3) Run the PCR instrument: Place the centrifuge tube on the heating module of the PCR instrument and start the PCR reaction.

[0078] The electrophoresis detection steps include:

[0079] (1) Prepare agarose gel: Prepare agarose gel of appropriate concentration according to the size of the DNA fragments to be separated, heat it to melt, cool it to an appropriate temperature, pour it into the electrophoresis tank, and insert a comb to form spot holes;

[0080] (2) Prepare the electrophoresis tank: Add electrophoresis buffer to the tank, make sure the gel is covered, and carefully remove the comb to prepare for sample spotting;

[0081] (3) Spotting: Mix the PCR product with the loading buffer and carefully add it to the spotting well. At the same time, add a DNA standard reference of known molecular size.

[0082] (4) Electrophoresis: Turn on the power supply and set the constant voltage for electrophoresis, such as 8V per cm of electrophoresis tank. Determine whether to terminate the electrophoresis based on the position of bromophenol blue.

[0083] (5) Observation results: Take out the gel and place it in a UV imager for observation. Determine the fragment size by comparing it with the DNA standard reference and record the experimental results.

[0084] Step S4: If the DNA quality meets the standard, the result is fed back to the classification neural network model for further training and optimization. If the DNA quality does not meet the standard, the automated workstation automatically adjusts the extraction parameters according to the strategy network, such as changing the buffer composition, adjusting the extraction time, or re-extracting. At the same time, the data in the database and the classification neural network model parameters are regularly updated.

[0085] The specific steps of step S2 include:

[0086] S2.1: Obtaining physical evidence sample data using feature concatenation ,in, Represents the physical property data of skin cells, Indicates the degree of skin cell pollution. Displays the storage condition information of skin cells. Represents the parameter annotation data of skin cells, represents the DNA spectrum data of skin cells, a, b, c, d, and e represent the physical property data of skin cells, the pollution degree information, the storage condition information, the parameter annotation data, and the quantity of DNA spectrum data, respectively;

[0087] S2.2: Select the pre-trained classification neural network model, take the physical evidence sample data and corresponding labels as input, and use the incremental learning strategy to optimize and update the parameters of the classification neural network model. The specific formula of the incremental learning strategy is:

[0088]

[0089] in, and represents the updated weights and biases of the lth layer, and represents the weight and bias of the lth layer, represents the error term of the lth layer, represents the modulation factor of the lth layer, represents the weight matrix, represents the weighted output of the lth layer, represents the derivative of the activation function, represents the learning rate, , They represent the momentum hyperparameters at the tth iteration in the Adam optimizer, represents the updated momentum term, represents the coefficient of the L2 regularization term, It represents the error term of the ith data in the physical evidence sample data X at the lth level, represents the Hadamard product, t represents the current iteration number, and m represents the number of data in the physical evidence sample data.

[0090] Furthermore, the specific steps of S2.2 include:

[0091] (1) Loading pre-trained model: Load the pre-trained neural network model and set the initial parameters of the model;

[0092] (2) Data preprocessing: Perform necessary preprocessing on the physical evidence sample data, such as normalization, standardization, and enhancement, so that the model can better process this data;

[0093] (3) Set training parameters: set learning rate, batch size, and number of training rounds hyperparameters;

[0094] (4) Construct a training loop: In the training loop, each iteration performs the following steps:

[0095] 1) Get a batch of sample data and corresponding labels from the dataset;

[0096] 2) Input the sample data into the model, perform forward propagation, and obtain the prediction results;

[0097] 3) Calculate the loss between the predicted result and the true label;

[0098] 4) Use the back-propagation algorithm to calculate the gradient of the loss to the model parameters;

[0099] 5) Update the model parameters using the gradient descent algorithm.

[0100] (5) Verify model performance: During the training process, the validation set is regularly used to evaluate the performance of the model so that hyperparameters can be adjusted or training can be stopped in a timely manner.

[0101] (6) Save the model: When the model performance meets the requirements, save the model parameters and structure.

[0102] S2.3: Evaluate the performance of the classification neural network model by calculating the accuracy, and adjust the structure, parameters or data preprocessing methods of the classification neural network model based on the evaluation results.

[0103] The specific formula of the policy network in step S3 is:

[0104]

[0105] in, represents the policy gradient, represents the parameter set of the nth policy network, n represents the number of policy network parameter sets, represents the state distribution under the strategy, represents the conditional probability, represents the original advantage function, h represents the action, s represents the state, represents the policy network, represents the distribution of context information k under a given state s, represents the weight of the jth parameter set, represents the hyperparameter, represents the reward function, represents the expected function, represents the gradient, Represents a combination set.

[0106] The specific formula for quality detection in step S3 is:

[0107]

[0108] in, Indicates DNA quality, Indicates the number of DNA copies at the beginning of the PCR reaction. represents the molecular weight of a single DNA copy, which depends on the length and base composition of the DNA fragment. q represents the amplification efficiency, which refers to the multiple of DNA fragment amplification in each PCR cycle. , u represents the number of cycles in the PCR reaction, represents the recovery efficiency of PCR products in subsequent processing, and , represents the purity factor, which indicates the impact of impurities in the PCR product, such as primer dimers and nonspecific amplification products, on the final DNA quality, and , when the purity of the PCR product is high, it is close to 1, and when the purity is low, it is less than 1.

[0109] Example 2

[0110] See also Figure 4 ,Another embodiment provided by the present invention: A sample automatic processing system based on a neural network, comprising: an image processing module, a control module, an evaluation module, and an interaction module;

[0111] An image processing module is used to preprocess, analyze and extract features from the image data of the sample;

[0112] A control module, which is used to automatically adjust the parameters of the processing equipment according to the outputs of the image processing module and the evaluation module, and perform the actual operation of sample processing;

[0113] The evaluation module is used to perform quality inspection on the processed physical evidence samples, perform performance evaluation on the classification neural network model, and feed back the quality inspection results and performance evaluation results to the control module and the image processing module to adjust the control strategy or model parameters;

[0114] The interactive module is used to provide an interface between the user and the system, allowing the user to input instructions, view processing progress, receive system feedback, and perform necessary manual intervention and parameter adjustments. In addition, it also includes alarm and notification functions to ensure that operators can respond to changes in system status in a timely manner.

[0115] The image processing module includes: image acquisition unit, image processing unit, feature extraction unit, process monitoring unit, data fusion unit, and model training unit;

[0116] An image acquisition unit, used for acquiring image data of forensic evidence samples;

[0117] An image processing unit is used to pre-process the collected image data, including noise filtering, graying, contrast enhancement, and size adjustment, so as to improve image quality;

[0118] A feature extraction unit is used to extract key features from the preprocessed image data using a neural network method, including the location, shape, and texture of the target object;

[0119] Process monitoring unit, which is used to track changes in the process of physical evidence sample processing in real time through image monitoring to ensure that the operation is carried out as expected and provide a basis for feedback control;

[0120] Data fusion unit, used to combine image features with spectral data and parameter annotation data to form a unified feature vector and prepare data for subsequent model input;

[0121] The model training unit is used to train a classification neural network model using historical data and feature vectors to predict optimal processing parameters or evaluate sample quality.

[0122] The control module includes: model embedding unit, adaptive control unit, decision making unit, and optimization unit;

[0123] Model embedding unit, used to integrate the trained neural network model into the automation system as the core engine of intelligent decision-making;

[0124] Adaptive control unit, used to dynamically adjust the parameters of the processing equipment, such as temperature, pressure, and time, according to the output of the classification neural network model, to achieve precise control of the processing process;

[0125] The decision-making unit is used to formulate the next operation instructions based on model prediction and real-time monitoring data, such as whether to continue the current processing flow, whether to adjust the control strategy, and whether to take action;

[0126] The optimization unit is used to collect processing results, optimize the control strategy and model parameters through the optimization algorithm, and improve the overall performance and processing efficiency of the system. The control strategy includes: dynamic parameter adjustment, process control, and exception handling.

[0127] The evaluation module includes: DNA analysis unit, quality assessment unit, database management unit, and model update unit;

[0128] DNA analysis unit, used to perform quantitative analysis of DNA fragments and assess extraction quality and integrity;

[0129] The quality assessment unit is used to evaluate the quality of the entire processing process based on the DNA analysis results to determine whether it meets the expected standards;

[0130] Database management unit, used to store processing results, sample data, processing parameters and model parameters, and provide support for data analysis, historical tracing and model iteration;

[0131] The model updating unit is used to update the model regularly or based on the evaluation results, including adjusting the model architecture and retraining the model to adapt to the latest data and improve the prediction ability.

[0132] Example 3

[0133] An electronic device includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of a method for automatic sample processing based on a neural network are implemented. For details, please refer to the above method embodiment, which will not be repeated here.

[0134] A computer-readable storage medium stores computer instructions, which, when executed, execute the steps of a neural network-based automatic sample processing method, wherein the storage medium may be a volatile or non-volatile computer-readable storage medium.

[0135] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation modes, which are merely illustrative rather than restrictive. Under the guidance of the present invention, ordinary technicians in the field may also change, modify, replace and modify the above-mentioned embodiments without departing from the purpose and scope of protection of the present invention, and all of these are within the protection of the present invention.

[0136] If the disclosed technical solution involves personal information, the product using the disclosed technical solution has clearly informed the personal information processing rules and obtained the individual's voluntary consent before processing the personal information. If the disclosed technical solution involves sensitive personal information, the product using the disclosed technical solution has obtained the individual's separate consent before processing the sensitive personal information, and at the same time meets the "explicit consent" requirement. For example, on personal information collection devices such as cameras, clear and prominent signs are set to inform that the personal information collection scope has been entered and personal information will be collected. If the individual voluntarily enters the collection scope, it is deemed that he or she agrees to the collection of his or her personal information; or on the device that processes personal information, when the personal information processing rules are notified by obvious signs / information, the individual's authorization is obtained through pop-up information or by asking the individual to upload his or her personal information; among them, the personal information processing rules may include information such as the personal information processor, the purpose of personal information processing, the processing method, and the type of personal information processed.

Claims

1. A sample automatic processing method based on a neural network, characterized in that: include: Step S1: Capturing a forensic evidence sample image by a scanner, extracting physical property data, contamination degree information, evidence preservation condition information, parameter annotation data, and DNA spectrum data of skin cells in the forensic evidence sample according to the forensic evidence sample image, and performing multimodal fusion to generate evidence sample data; Step S2: input the physical evidence sample data into the pre-trained classification neural network model for training, and optimize the classification neural network model parameters using the incremental learning strategy to obtain the optimized classification neural network model; Step S3: embed the optimized classification neural network model into the control system of the automated workstation, and train a policy network in combination with reinforcement learning. After the DNA extraction is completed, the extracted DNA is quality tested using PCR amplification and electrophoresis methods; Step S4: If the DNA quality meets the standard, the result is fed back to the classification neural network model for further training and optimization. If the DNA quality does not meet the standard, the automated workstation automatically adjusts the extraction parameters or re-extracts according to the strategy network. At the same time, the data in the database and the classification neural network model parameters are regularly updated. The specific steps of step S2 include: S2.1: Obtaining physical evidence sample data using feature concatenation ,in, Represents the physical property data of skin cells, Indicates the degree of skin cell pollution. Displays the storage condition information of skin cells, Represents the parameter annotation data of skin cells, represents the DNA spectrum data of skin cells, a, b, c, d, and e represent the physical property data of skin cells, the pollution degree information, the storage condition information, the parameter annotation data, and the quantity of DNA spectrum data, respectively; S2.2: Select a pre-trained classification neural network model, take the physical evidence sample data and corresponding labels as input, and use an incremental learning strategy to optimize and update the parameters of the classification neural network model; S2.3: Evaluate the performance of the classification neural network model by calculating the accuracy, and adjust the structure, parameters or data preprocessing method of the classification neural network model according to the evaluation results; The specific formula of the incremental learning strategy in S2.2 is: ； in, and represents the updated weights and biases of the lth layer, and represents the weight and bias of the lth layer, represents the error term of the lth layer, represents the modulation factor of the lth layer, represents the weight matrix, represents the weighted output of the lth layer, represents the derivative of the activation function, represents the learning rate, , They represent the momentum hyperparameters at the tth iteration in the Adam optimizer, represents the updated momentum term, represents the coefficient of the L2 regularization term, It represents the error term of the ith data in the physical evidence sample data X at the lth level, represents the Hadamard product, t represents the current iteration number, and m represents the number of data in the physical evidence sample data; The specific formula of the policy network in step S3 is: ； in, represents the policy gradient, represents the parameter set of the nth policy network, n represents the number of policy network parameter sets, represents the state distribution under the strategy, represents the original advantage function, h represents the action, s represents the state, represents the policy network, represents the distribution of context information k under a given state s, represents the weight of the jth parameter set, represents the hyperparameter, represents the reward function, represents the expected function, represents the gradient, Represents a combination set; The specific formula for quality detection in step S3 is: ； in, Indicates DNA quality, Indicates the number of DNA copies at the beginning of the PCR reaction. represents the molecular weight of a single DNA copy, q represents the amplification efficiency, and u represents the number of cycles in the PCR reaction. represents the recovery efficiency of PCR products in subsequent processing, and , Represents the purity factor.

2. The method for automatic sample processing based on a neural network as claimed in claim 1, characterized in that: The physical property data of the skin cells is obtained by performing target detection on the forensic evidence sample image based on the improved convolutional neural network, locating the position of the forensic evidence sample in the image, and obtaining it according to the target detection and positioning information; the pollution degree information is obtained by separating the DNA sample area through semantic segmentation, identifying the pollutant area, and evaluating the preservation state of the skin cells in the pollutant area, and then obtaining the pollution degree of the skin cells according to the evaluation results; the evidence preservation condition information is obtained according to the environment where the skin cells are located; the parameter annotation data is obtained by using an automated workstation combined with a real-time fluorescence quantitative method to extract DNA from skin cells, monitor the DNA extraction process, and automatically record and annotate the process parameters of the DNA extraction to form parameter annotation data; The DNA spectrum data is obtained by performing spectrum analysis on the extracted DNA to obtain the DNA spectrum data.

3. A neural network-based sample automatic processing system, which is used to implement the neural network-based sample automatic processing method according to any one of claims 1 to 2, characterized in that: include: Image processing module, control module, evaluation module, interaction module; The image processing module is used to preprocess, analyze and extract features from the image data of the forensic evidence sample; The control module is used to automatically adjust the parameters of the processing equipment according to the outputs of the image processing module and the evaluation module, and perform the actual operation of physical evidence sample processing; The evaluation module is used to perform quality inspection on the processed physical evidence samples, perform performance evaluation on the classification neural network model, and feed back the quality inspection results and performance evaluation results to the control module and the image processing module to adjust the control strategy or model parameters; The interaction module is used to provide an interface between the user and the system, allowing the user to input instructions, view processing progress, and receive system feedback.

4. The automatic sample processing system based on neural network as claimed in claim 3, characterized in that: The control module includes: an adaptive control unit, a decision-making unit, and an optimization unit; The adaptive control unit is used to dynamically adjust the parameters of the processing device according to the output results of the classification neural network model; The decision-making unit is used to formulate the next operation instruction based on the model prediction and real-time monitoring data; The optimization unit is used to collect processing results and optimize the control strategy and model parameters through an optimization algorithm.

5. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the automatic sample processing method based on a neural network according to any one of claims 1 to 2 are implemented.

6. A computer-readable storage medium, characterized in that: Computer instructions are stored thereon, and when the computer instructions are executed, the steps of the automatic sample processing method based on a neural network as described in any one of claims 1-2 are executed.

Images