54 results about "Sample Type" patented technology

The invention discloses an electric arc detection method based on differential increase and structured attention, and the method comprises the steps: firstly carrying out the collection and preprocessing of a current signal, carrying out the differential enhancement according to a sample type, carrying out the strong enhancement of an electric arc sample, improving the generalization capability, and carrying out the weak enhancement of a normal sample, thereby avoiding the overfitting; the problem of class imbalance is relieved, and the model generalization ability is improved; secondly, obtaining six complementary feature representations of time domain waveform, frequency domain frequency spectrum, time frequency analysis, envelope features, statistical distribution and related features from the differentially enhanced current signal through a multi-modal feature extraction method, and fusing to generate a multi-modal image; then, designing a deep learning model integrated with structured attention, carrying out distinguished attention on different feature analysis areas of the multi-modal image, and directionally enhancing arc features; and finally, dynamically quantifying the trained model, reducing the size of the model and reasoning delay, and supporting efficient deployment of various edge computing devices.

The invention discloses a microbe grouping analysis method and related equipment. The method comprises the following steps: acquiring microbe data of a target research area; based on microorganism abundance data, a conversion value of each sample is obtained through center logarithm ratio conversion processing, and then a distance value between any samples is obtained through Euclidean distance quantization; adaptively setting a minimum sample number and a neighborhood radius based on a sample number and a species number corresponding to the microbial data; performing density clustering on samples in the microbial data to obtain a clustering result, and further marking the sample type of each sample in combination with the minimum sample number and the neighborhood radius; performing multi-level analysis on the clustering result based on the sample type to obtain an analysis result of the clustering result; and performing correlation analysis based on the analysis result and the methane leakage data to obtain a microbe grouping analysis result of the target research area. The method realizes standardization and automation of the analysis process, can remarkably improve the analysis efficiency, and can be widely applied to the technical field of data processing.

The invention discloses a multi-modal large model Deepfake detection method based on retrieval enhancement, and relates to the frontier crossing field of artificial intelligence security, computer vision and multi-modal information processing, and the method comprises the steps: 1, carrying out two-stage labeling in sequence, constructing an evidence obtaining knowledge database through the first-stage labeling, constructing an evidence obtaining thinking chain data set through the second-stage labeling, and carrying out the second-stage labeling; wherein the evidence obtaining thinking chain data set is used for generating thinking chain data including a cross validation type, an evidence guide correction type and an adversarial sample type according to the validity of initial judgment and retrieval evidences; step 2, for an image to be detected, retrieving text evidences of Top-k similar cases from the evidence obtaining knowledge database by using a dynamic evidence obtaining retriever; and step 3, inputting an image to be detected and the retrieved text evidence into the critical reasoning multi-modal large model, performing multi-modal reasoning analysis, and outputting a detection result containing a reasoning process.

The application discloses a target detection model evaluation system and method based on light weight, relates to the technical field of target detection, and ensures data specification by acquiring a data set and converting the data set into a YOLO format, divides sample types by matching scores, and dynamically adjusts weights by using a segmented function to optimize classification; subsequently, an SPPF-Attention unit is constructed by combining a 1*1 point-by-point convolution, a depth-by-depth convolution and an MHSA module, and a light weight model is built; samples are expanded by Mosaic data enhancement, and the training process is controlled by using a segmented function; finally, the performance of the model is evaluated from multiple dimensions of parameter quantity, calculation quantity, mAP value and inference time consumption, the organic unification of model light weight, high-precision detection and high-efficiency inference is realized, the training efficiency and generalization capability are improved, and the multiple requirements of actual application are met.

The invention relates to the technical field of production line management, and provides a full-automatic production line management method and system for phosphorus flame retardant pretreatment, and the method comprises the steps: building a resource demand model of a multi-medium pollution sample treatment task, and according to the type of a multi-medium pollution sample, a target treatment flow and preset technological parameters, obtaining a resource demand model of the multi-medium pollution sample treatment task; calculating an expected demand of each multi-medium pollution sample processing task on the shared key resource in a future preset duration; aggregating the expected demands of the multi-medium pollution sample processing tasks, and generating a whole production line future resource demand prediction map; obtaining shared resource availability data in real time, and comparing the shared resource availability data with a whole production line future resource demand prediction map and a preset resource supply period to identify a resource shortage risk; when a resource shortage risk is identified, making a resource allocation decision; and executing a resource allocation decision, and controlling the full-automatic production line. The full-automatic production line has the effect of improving the operation efficiency and stability of the full-automatic production line.

The application discloses a biomedical test strip detection data analysis method and system based on multiple algorithms. The method comprises the following steps: obtaining related data of a test paper to be detected, wherein the related data comprises T-line related data, B-line related data and C-line related data; verifying whether the lengths of three curves corresponding to the related data are consistent, if not, determining that the detection result is invalid, if yes, analyzing sample adding characteristics to determine whether there is sample adding sequence error or no sample adding; distinguishing sample types according to chromatography speed according to the related data; when the sample type is a whole blood type, calculating the length of a sample type identification point corresponding to three lines, adjusting and cutting off, and if the total length exceeds a set threshold, determining that a hemolysis phenomenon leads to an invalid detection result. The method of the application particularly focuses on processing the hemolysis phenomenon and other user operation related problems, thereby enhancing the reliability and efficiency of the overall detection system.

The invention discloses an entry sample gamma radionuclide passive efficiency calibration method and system, and the method comprises the steps: collecting entry sample nuclide distribution, detector response, measurement geometry and other basic parameters, and then calling a ray direct penetration rate prediction model to calculate penetration rate data, an optimized detector parameter set is generated through a detector parameter characterization optimization model, correlation mapping of energy spectrum data and efficiency scale parameters is completed through an entry nuclide energy spectrum data analysis platform, target nuclides are recognized through a nuclide type recognition matching algorithm, feature parameters are extracted, and finally passive efficiency scales are completed. According to the method, a collaborative linkage mechanism is formed by integrating multiple models, a systematized data processing link is established, scenes with complicated types of entry samples and non-uniform nuclide distribution are effectively adapted, the accuracy and consistency of scales can be guaranteed without depending on a standard source, and efficient and reliable technical support is provided for entry nuclide detection.

This application provides a liquid analyzer and its automatic pipette tip selection detection method and storage medium. The method includes: acquiring consumable information and detection task information; the consumable information includes pipette tip model, available quantity of pipette tips, mixing cup specifications, and available quantity of mixing cups; the detection task information includes the detection items to be performed and the sample type; determining the sample volume and sample processing liquid volume that meet the pipetting conditions based on the detection task information and the detection method corresponding to the detection items; obtaining the pipetting coefficient of variation based on the sample type, sample volume, and sample processing liquid volume; selecting a target pipette tip model that meets preset detection conditions based on the pipetting coefficient of variation and the consumable information; and selecting the corresponding target pipette tip to perform the detection task and obtain the detection result. This method can reduce detection errors caused by improper pipette tip selection and effectively improve the accuracy and efficiency of detection.

This application relates to a method and system for identifying defects in corundum-coated ground. The method constructs a first feature distribution and a second feature distribution based on image features from X-ray image samples. The first feature distribution refers to a comprehensive feature data distribution formed by fusion of valuable and conventional sample features through a weighted summation method. This balances the influence of different sample types on the feature space, and the combination of valuable and conventional samples can represent important information about the data distribution characteristics, accurately reflecting the data distribution. The second feature distribution represents the feature distribution of pixels. Combining the comprehensive expression of the two distributions and matching them in a database allows for the matching of a more accurate sample set of data distributions. Finally, based on the X-ray image samples and the matched sample set, the defect detection model is trained, improving the accuracy of the defect detection model training.

Embodiments of the present application provide a data processing method and device, and a prediction model training method and device. The method comprises: determining feature information according to target data and access records of the target data; inputting the feature information into a prediction model to obtain time information of future access of the target data; the prediction model is trained by using training samples, the training samples comprising sample features, sample time labels and sample types, the sample time labels and the sample types being determined according to whether there are access records of data corresponding to the sample features before and after a random time within a sample sampling period; and identifying cold and hot data of the target data according to the time information. The sample time labels are used as labels of the training samples to train the prediction model, and then the time interval of next access of the target data based on the prediction model is used as a prediction result, and cold and hot data of the target data is identified according to the time interval, so that the accuracy of cold and hot data identification of the target data can be effectively improved.

The invention relates to the technical field of artificial intelligence, in particular to a multi-agent cooperative detection method, system and equipment based on a knowledge graph, and the method comprises the steps: receiving detection task data through a preset detection sample management agent, and carrying out the semantic understanding and structural processing of the detection task data through a preset large language model, and based on the processing result, performing knowledge routing in a preset knowledge graph to identify a sample type and determine a detection process, performing a detection experiment through a preset detection sample experiment recording agent according to the detection process, and interacting with a database of the detection equipment through a multi-modal data acquisition technology to acquire and record experiment data; and calling the knowledge graph through a preset report generation agent to perform automatic comparison on the experimental data, and performing intelligent judgment under the assistance of the large language model so as to generate a detection report. In this way, the automation level, efficiency and accuracy of detection can be remarkably improved.

The utility model relates to the technical field of vehicle inspection station sample detection, in particular to an equipment sample recognition and conveying mechanism for a vehicle inspection station, and the equipment sample recognition and conveying mechanism comprises a conveyor belt main body which is provided with a conveyor belt and a driving motor, the conveyor belt is in transmission connection with the driving motor, and the driving motor drives the conveyor belt to move when working; the sample identification camera is arranged on the adjacent side of the conveying belt and is used for identifying a sample image on the conveying belt and converting the sample image into a sample type signal to be output; and the sorting mechanical arm is arranged adjacent to the sample identification camera and is used for receiving the sample type signal sent by the sample identification camera and sorting the samples. The device has the effect of improving the working efficiency of sample detection of equipment.

This utility model discloses a sample label sorting device for a laboratory, including a storage box; it further includes: a sliding groove on one side of the bottom of the storage box, a positioning tooth fixedly connected to the bottom surface of the storage box near the sliding groove, a sliding rod slidably connected to the middle of the sliding groove, a return spring sleeved on the outer side of the sliding rod, a sliding sleeve sleeved on the outer side of the return spring, and a locking tooth fixedly connected to the top of the sliding sleeve; wherein, a separator is fixedly connected to the top of the sliding rod, a limit rod is provided on one side of the top of the separator, a clamping spring sleeved on the outer side of the limit rod, and a clamping plate is attached to one end of the clamping spring. The separator can be moved by pulling, which facilitates the classification and storage of labels according to sample type or test item, and the position of the clamping plate can be adjusted by sliding, which facilitates the clamping of sample labels and thus prevents label confusion.

This disclosure describes a method for determining, in real time and continuously, an amount of particles of a given material in a sample, including steps of: E1: optically measuring a signature of the sample by an optical sensor; E2: identifying, by a processing unit, the type of the sample by a classification model trained on a training database including a plurality of reference signatures, each reference signature being associated with a reference sample type; and E3: determining, by the processing unit, the amount of particles of the given material in the sample based on the identified sample type and a correspondence table associating, with each of a plurality of reference sample types, a reference amount of particles of the given material.

The present application relates to the technical field of rail processing, and discloses an automatic processing method for rail samples, which comprises the following steps: obtaining image information and label information of a rail to be processed; obtaining a sample type to be processed, and calling a corresponding processing model from a preset processing model database in combination with the label information; automatically clamping and positioning the rail to be processed based on the processing model and the image information; sequentially performing bulk sample cutting, small sample decomposition and finishing on the clamped and positioned rail blank based on the processing model, so as to obtain a finished sample; and marking the finished sample and automatically discharging it. The method solves the problems of low processing efficiency, poor precision consistency, insufficient intelligence, and high safety risks caused by relying on manual operation in the processing of existing rail samples.

The invention discloses a remote sensing image labeling system and method. A basic support module of the labeling system provides visualization and editing functions of image and vector data and project management and system database management functions; the standard labeling process module provides a unified labeling process for various labeling types; the sample batch generation and conversion module provides various forms of sample batch generation and conversion functions; the labeling method comprises the following steps: firstly, creating a labeling project, setting a classification system and a standard labeling process mode, determining a standard labeling process according to the selection of the classification system and the selection of the labeling mode, and completing project construction and initializing the project by the system according to the selection; the system can carry out semi-automatic labeling, and sample batch generation and format conversion can be realized. According to the invention, sample data required by remote sensing image deep learning is efficiently manufactured in a standardized manner, all sample types of current remote sensing deep learning are covered, and the system and the method are a full-stack system and method.

The invention discloses a sample data centralized optimization processing method applied to image recognition, and relates to the technical field of sample image data processing. According to the method, visual features of error sample data are extracted, clusters are obtained through a clustering algorithm, and cluster attributes are analyzed to determine gap sample types; calling an association model corresponding to the model task, matching a sample quality dimension corresponding to a model abnormal index, and screening out an abnormal target quality dimension through a pre-training model or manual verification of a sample data set; finally, a sample data set of the target model is optimized according to the logic that quality optimization is conducted firstly and then quantity optimization is conducted; the method breaks through the limitation that data requirements are judged according to experience in traditional static data set processing, on one hand, gap scenes which are not covered by a model are accurately positioned through clustering, and the problem that samples are redundant or key scene samples are insufficient is avoided; and on the other hand, low-quality data is removed in combination with the target quality dimension, and it is ensured that the supplemented samples have both the number standard and the quality adaptation.

The invention discloses a mutation back noise filtering algorithm based on next-generation sequencing data, relates to the technical field of high-throughput sequencing data analysis, and aims to solve the problems that the prior art depends on a baseline, and is poor in adaptability and high in false positive. The algorithm comprises the following steps: preprocessing sequencing data to obtain a pileup file, classifying sites according to a sequence context, calculating an error rate, constructing a layered back noise model to check mutation, qualitatively determining real mutation and back noise by combining a sample type, and evaluating sample quality. The method does not need to construct a base line in advance, adapt to hybrid capture and amplicon sequencing, can improve mutation detection specificity and repeated sample consistency, guarantees result reliability, and is suitable for scenes such as tumor gene detection.

The embodiment of the invention provides instant detection equipment. The instant detection equipment comprises a scanning assembly, an incubation assembly, a detection assembly and a processor, the scanning assembly is used for acquiring a detection item of a sample, the incubation assembly is used for incubating the sample, and the detection assembly is used for detecting the incubated sample to obtain a detection signal; the processor is used for controlling the incubation assembly to incubate the sample according to a detection item of the sample, and controlling the detection assembly to detect the sample to obtain a detection signal of the sample no matter whether the sample type of the sample is obtained or not after the incubation of the sample is finished; and obtaining a sample type of the sample set by the user, and determining a detection result of the sample according to the sample type and the detection signal of the sample. The instant detection equipment provided by the embodiment of the invention is relatively high in detection efficiency, and can prevent the situation that the sample incubation time is too long or the incubated sample is stored too long when the sample incubation is finished, so that the detection result of the sample is inaccurate.

The application provides a non-standard rock sample compressive strength correction method, comprising: collecting rock samples generated under different formation conditions from different oil and gas blocks, measuring the height and diameter of each sample, determining the corresponding sample type, and simultaneously obtaining the lithology of each sample, and measuring the compressive strength of each sample by setting a conventional experiment; the height, diameter, confining pressure, and compressive strength are dimensionally standardized, and the sample data is randomly divided into a training set and a test set with consistent sample type and lithology distribution according to a ratio of 7:3; a quantile regression model based on GBDT is constructed using the training set, the hyperparameters of the quantile regression model are optimized using a particle swarm optimization algorithm, the test set is used to verify the quantile regression model to output the compressive strength correction values of multiple key quantiles, and the uncertainty range of the correction result is quantified. The method can accurately establish a unified standard strength benchmark that adapts to different height-diameter ratios and different lithologies, and greatly improves the correction accuracy.

According to a first aspect of the present disclosure, the present disclosure describes a computer-implemented method for determining a particular analysis protocol for a sample, the sample being one of a plurality of sample types, and each sample type being associated with a corresponding particular analysis protocol. Performing the method comprises: obtaining a baseline spectrum of the sample using a baseline analysis scheme, the baseline analysis scheme being the same for the plurality of sample types; providing a machine learning model, such as a convolutional neural network, trained to output output data indicative of a particular analysis scheme for a sample in response to a spectrum of the sample; taking the obtained spectrum of the sample as the input of the machine learning model, and obtaining the output of the machine learning model; and determining a specific analysis protocol for the sample based on the output. This minimizes the burden on determining the correct analysis protocol for the sample.

The invention relates to the technical field of sewer defect detection, in particular to a sewer defect detection method based on YOLO11 and DDAA, and the method comprises the steps: obtaining marked sewer defects, measuring the length and width, calculating the area, setting three thresholds for comparison according to the area, dividing tiny, fine, medium and larger defect intervals, selecting tiny defect samples, extracting features by DDAA, evaluating the optimal expansion rate, and determining each interval value. A matching table is generated, training set defects are extracted and screened according to the table, types are manually marked, features are calculated to obtain a corresponding table, a crack ratio, a length-width ratio and a threshold value in the table are stretched, rules are determined according to the same principle for other defects, and a morphological enhancement matching result is obtained. The method comprises the following steps: acquiring the length and width of a marked defect, calculating the area, setting a threshold value according to the area, dividing size intervals, selecting samples in each interval, extracting features, determining an expansion rate, screening sample types according to the intervals, calculating morphological parameters, determining an enhancement rule according to the comparison of the parameters and the threshold value, estimating a potential defect area through feature extraction, adjusting the expansion rate, and processing training data in combination with the rule. And feature extraction is adaptive to defects of different sizes.

A sample slide preparation device and method are disclosed, wherein the sample transport device transports samples including trace blood samples and normal blood samples. A trace blood sample mixing device drives the movement of the trace blood sample to mix it. A normal blood sample mixing device drives the movement of the normal blood sample to mix it. A control unit determines whether the sample slide preparation device operates in normal blood mode or trace blood mode based on sample type information, and selects different methods to complete sample mixing, sample collection, and movement according to the different modes, finally preparing a sample slide for subsequent detection and analysis.

A sample analyzer (1) and a sample analyzer method, the sample analyzer (1) comprising: a sample container accommodating device (90) configured to accommodate a sample container (91, 92, 93) loaded with a sample (100); a capacitance sensor (7) configured to detect a sample amount or a sample position of the sample (100) in the sample container (91, 92, 93) in a non-contact manner with the sample (100); a sample processing device (50) configured to process the sample (100) in the sample container (91, 92, 93); and a control device (30) communicatively connected with the capacitance sensor (7) and the sample processing device (50) and configured to: acquire sample information of the sample container (91, 92, 93) from the capacitance sensor (7), the sample information comprising at least one of sample amount information and sample position information, and control a processing action of the sample processing device (50) or determine whether the sample processing device (50) implements the processing action according to the sample information. Thus, the sample container type or the sample type can be reliably identified according to the capacitance sensor (7), and the subsequent processing action of the sample (100) in the sample container (91, 92, 93) is determined, thereby improving the safety of sample analysis.

The invention discloses a heterogeneous federated learning method and system based on personalized collaborative generation, belongs to the technical field of crossing of federated learning and generative models, and aims to solve the core problems of coexistence of data and model heterogeneity, dependence on a common data set and poor undersampling learning effect in existing federated learning. According to the method, efficient and privacy-protected personalized training is realized through a two-stage collaborative framework, each client trains a conditional variation auto-encoder to capture local data distribution, and a priori distribution offset mechanism of privacy enhancement is adopted; in addition, a uniform feature representation space is constructed by selecting a lens space similarity graph through a reference encoder, cross-model knowledge migration and under-sampling type pertinence enhancement are realized by combining collaborative knowledge updating, matching of auxiliary generators driven by class recognition capability and self-adaptive synthetic sample generation, and finally, a local model is optimized by adopting mixed loss. The method does not need to depend on a public data set, takes performance, privacy and deployment flexibility into consideration, and is suitable for privacy sensitive fields such as edge device cooperative training, medical treatment and finance.

The application discloses an anchor weight distribution optimization method, equipment and medium, wherein the method comprises the following steps: collecting corresponding pictures according to a to-be-detected target and performing detection labeling; setting a detection model and an anchor, inputting the pictures and the detection labeling; calculating the iou of each anchor and all gtboxes of each picture; determining the sample type of each anchor of each picture; counting the number of each sample and total gtboxes in the batch of pictures; setting the weight of classification and regression loss of each anchor of each picture; performing normalization or weight adjustment again according to a traditional additional loss weight normalization method; calculating the final total loss according to the set target value and weight of classification and regression; and obtaining a final detection model for detecting the to-be-detected target. The application can significantly improve the detection effect of objects with extreme position, size and width-height ratio.

The application discloses an arc detection method based on differential increase and structured attention. Firstly, current signal acquisition and preprocessing are performed, and differential enhancement is implemented according to sample types. Strong enhancement is adopted for arc samples to improve generalization ability, and weak enhancement is adopted for normal samples to avoid overfitting, thereby relieving the class imbalance problem and improving the model generalization ability. Secondly, six complementary feature representations, including time domain waveform, frequency domain spectrum, time-frequency analysis, envelope feature, statistical distribution and correlation characteristics, are obtained from the differentially enhanced current signal through a multi-modal feature extraction method, and a multi-modal image is fused and generated. Then, a deep learning model integrating structured attention is designed to distinguishably focus on different feature analysis regions of the multi-modal image and directionally enhance arc features. Finally, the trained model is dynamically quantized to reduce the model size and inference delay, and support efficient deployment of various edge computing devices.

The present application belongs to the technical field of metallurgical quality detection, and relates to a method for tracking steel sample detection time, comprising the following steps: S1, recording sample information in the sample detection process, including sample type, sample code, order arrival time, sample sending time and sample arrival time; S2, printing a sample label through a label printer and sticking the label on the sample; S3, sending the sample to a preparation station for sample preparation; S4, scanning a two-dimensional code at the sample code column of a direct-reading spectrometer to input the sample code; and S5, displaying sample composition data and sample analysis time. According to the present application, the detection time of each sample is dynamically and real-timely displayed in the detection and analysis system of the steel sample, so that the problem that the operation personnel cannot intuitively understand the detection time and cannot timely and accurately fill in the reason for exceeding the detection time is solved, thereby avoiding the situation that the detection result is affected by untimely information transmission and it is difficult to trace the failure reason.

The invention discloses an adaptive method during open world testing based on a hierarchical ladder network. The method comprises the following steps: inputting each target domain image into a pre-trained source domain model to obtain a first category prediction result and an entropy value; determining a sample type of each target domain image in the current batch according to the entropy value; if the sample type of the target domain image is an out-of-distribution sample, generating a second category prediction result based on a hierarchical ladder network set in the source domain model and the category mark of each layer of encoder; fusing the first category prediction result and the second category prediction result to obtain an out-of-distribution discrimination score; network parameters of the hierarchical ladder network are updated through back propagation, and a target domain model is obtained; and determining the sample type of the input current target domain image based on the hierarchical ladder network in the target domain model, and outputting a classification result. The problem that the prediction accuracy of the adjusted model is low due to the fact that the model is prone to misleading after the samples outside the distribution and the samples inside the distribution are mixed is solved.

The invention belongs to the technical field of artificial intelligence, and discloses a multi-modal large model input attack detection method and a related device, and the method comprises the steps: obtaining a to-be-detected input sample of a multi-modal large model; carrying out variation on the to-be-detected input sample based on a preset sample variator to obtain a plurality of variation samples; obtaining the model response of each variation sample based on the multi-modal large model, obtaining the maximum difference between the model responses of different variation samples, and obtaining a to-be-analyzed difference value of the to-be-detected input sample; and obtaining a sample type of the to-be-detected input sample according to the to-be-analyzed difference value of the to-be-detected input sample in combination with a preset difference value threshold value. Whether the input sample is the attack sample is effectively detected based on the difference of the variation samples, the robustness feature design based on the input attack is not only suitable for the attack sample constructed by a specific method, the generalization ability is high, multiple attack modes can be effectively detected, and the detection method is easy to implement, low in complexity and high in universality.