52results about How to "Avoid noise disturbance" patented technology

The invention relates to the technical field of particle detection, in particular to a SiPM test and calibration method and system for particle detection application, and the method comprises the following steps: extracting a slope quadratic difference set to determine a current change inflection point, calculating a local slope change proportion set to obtain a breakdown breakover voltage point, analyzing the output pulse triggering probability, and calculating the current change inflection point; and in combination with a photon number extraction probability growth interval, constructing a refined pulse extraction detection efficiency reference ratio, screening high-density trigger data, and calculating a spatial deviation extreme value, so that the system bias compensation voltage is adjusted. According to the method, the breakdown breakover voltage point is positioned, so that measurement interference caused by basic background noise is avoided, the attenuation number is counted to establish a trigger probability sequence, a preset maximum likelihood estimation model is combined to extract a detection efficiency reference ratio, a channel deviation extreme value is calculated to directly correct parameter fluctuation, and array global dynamic balance is established. And the long-term stability of the whole detection working state is maintained.

The invention discloses a code abstract generation method and system based on hierarchical context awareness, and belongs to the field of natural language processing and software engineering. The method comprises the following steps: generating hierarchical annotations for a code warehouse based on a function call graph, firstly extracting a call relationship in the graph and eliminating a ring structure, and then classifying leaf nodes and generating annotations; a topological sequence from bottom to top is adopted in the core step, the importance score of a child node is calculated for each father node, child node annotations, class function descriptions and code bodies of the child nodes are fused, and a large language model is input to generate annotations till the whole graph is covered. The method is suitable for an automatic document generation scene of a complex software project, and through a technical path combining function call graph analysis, child node importance measurement and class function description, the accuracy and semantic consistency of code annotation are improved; the method can be widely applied to the fields of software maintenance, code understanding, intelligent development tools, automatic document generation of enterprise-level code warehouses and the like.

The application discloses a colorectal cancer metastasis evaluation method based on NF-kappa B pathway activity, relates to the technical field of medical treatment, and comprises the following steps: obtaining a colorectal cancer tissue sample and a paired peripheral blood sample, separating exosomes, and dividing the tissue sample into a spatial resolution sub-sample and a function verification sub-sample; performing multiple immunofluorescence imaging on the spatial resolution sub-sample, identifying a phosphorylated p65 positive cancer cell group and a phosphorylated STAT3 positive immune cell group, and extracting cell phenotype data; calculating the spatial interaction weight value between high-activity cells based on the cell phenotype data to obtain a tumor microenvironment paracrine activation index; detecting the abundance of an upstream phosphorylated protein of the NF-kappa B pathway and the expression amount of a downstream metastasis effect gene in the function verification sub-sample, and obtaining a pathway function coupling score in combination with the matching degree of the two; and determining the content of a phosphorylated IKK protein and the content of a specific metastasis-promoting microRNA in the exosomes, and obtaining an exosome double-lock activity score by using the product relationship between the two.

The application discloses a kind of pathological image classification methods based on dyeing perception intelligent feature modeling, belong to pathological image classification field, this method includes collecting the multi-staining full section image group data of different objects;Multi-staining full section image group data includes H&E dyeing image and several immunohistochemical dyeing images;Each multi-staining full section image group data is respectively preprocessed, obtains the several equal-size image blocks of each object;Based on the image block of each object, graph is constructed, and the multi-modal heterogeneous pathology graph of each object is obtained;With the multi-modal heterogeneous pathology graph of each object as training data, train pathological image classification model;The pathological image classification model of training completion is used to classify the pathological image to be measured.The application solves the problems of the existing method in processing digital pathology graph data, such as node category characteristics, edge information utilization, cross-staining interaction and non-perfect data processing.

The application discloses a Mark point coordinate detection method based on an image sequence and a related device, and is used for improving display screen Mark point positioning precision. Image acquisition of N frames in succession is performed on a to-be-positioned object provided with a Mark point; Mark point coordinates of a photographed image sequence are extracted to generate a candidate coordinate point set; according to each candidate coordinate point, a gray feature of a photographed image where the candidate coordinate point is located is extracted to generate a gray feature set; a position deviation amount of each candidate coordinate point is calculated according to the candidate coordinate point set to generate a position deviation amount set; a gray consistency parameter of each candidate coordinate point is calculated according to the gray feature set; a multi-dimension confidence index of each candidate coordinate point in the candidate coordinate point set is calculated according to the position deviation amount set and the gray consistency parameter set; the candidate coordinate point set is screened according to the multi-dimension confidence index set, and a target coordinate of the Mark point is generated by using a screened candidate coordinate point subset.

The invention discloses a high-capacity circuit breaker heat dissipation structure based on a micro-nano structure heat pipe, and belongs to the technical field of circuit breaker heat dissipation. Comprising a vacuum arc extinguish chamber, an arc extinguish chamber heat dissipation device, a micro-nano structure heat pipe and a heat pipe heat dissipation device. The arc extinguish chamber heat dissipation devices are arranged on the two sides of the vacuum arc extinguish chamber, the micro-nano structure heat pipe is embedded in a static conducting rod of the vacuum arc extinguish chamber, and the heat pipe heat dissipation fins are arranged on the heat pipe. Through cooperation of the arc extinguish chamber heat dissipation device and the micro-nano structure heat pipe, radial conduction and axial rapid conduction of heat of the vacuum arc extinguish chamber are realized, and temperature rise of the high-capacity circuit breaker is remarkably reduced.

The application provides a Chinese-oriented named entity recognition method and system, and belongs to the technical field of natural language processing; comprising: an encoding stage step, context encoding of input Chinese text; a feature extraction stage, a double-branch structure of global semantic branch and local boundary branch in parallel is adopted, wherein the local boundary branch carries out boundary feature enhancement through multi-granularity convolution combined with a learnable Gaussian Laplacian operator; a feature fusion and prediction stage step, global and local features are fused through a gating mechanism, and multi-label classification prediction is carried out based on a preset six-tuple grid label system; a decoding stage step, named entities are obtained through bidirectional search strategy decoding according to predicted relationship probability; through boundary perception feature extraction and enhanced label system, the recognition accuracy and recall rate of nested entities and discontinuous entities in Chinese text are significantly improved.

The invention relates to the technical field of sensors, in particular to a multi-dimensional force sensor which comprises a shell, an elastic body, a supporting structure and a plurality of dampers. The elastic body is arranged in the shell and comprises a load input end for receiving an external load and a mounting part connected with the shell, and the elastic body is used for generating deformation in response to the external load; the supporting structure is arranged on the periphery of the elastic body in a sleeving mode and located in the shell, and the supporting structure is connected with the load input end; the multiple dampers are arranged between the supporting structure and the shell; wherein the damper is configured to dissipate the vibration energy of the elastic body through a viscous dissipation effect or flexible deformation. According to the multi-dimensional force sensor provided by the invention, the dynamic performance can be directly and effectively improved from a mechanical structure, the self vibration is inhibited, and the response time is shortened.

The application provides a kind of tool notch detection method, device, equipment and readable storage medium, applied to defect detection technical field.The edge of rotating tool is detected in real time by using photoelectric sensor, wherein the edge of tool at least shields part of detection light path of photoelectric sensor, and tool detection signal output by photoelectric sensor is acquired in real time.The light flux of detection light path not being shielded by tool is positively correlated with tool detection signal output by photoelectric sensor, and the time corresponding to one rotation of tool is used as a period, the maximum extreme point and minimum extreme point in tool detection signal waveform are found in each period, and finally whether tool exists notch is determined according to the slope of line between maximum extreme point and minimum extreme point.It can be seen that the tool in rotating working state can be detected in real time by photoelectric sensor, whether tool exists notch can be quickly judged, and the accuracy of tool damage notch detection result is improved.

This application discloses a method, apparatus, and equipment for adjusting navigation tasks based on human intelligence function allocation, belonging to the field of ship control technology. The method includes: performing cross-band dynamic functional connectivity analysis on physiological signals to extract neural state trajectories; performing task semantic parsing on navigation operation logs to extract behavioral log streams corresponding to the ship's sub-task types; constructing positive and negative sample pairs based on the synchronization state of the neural state trajectories and behavioral log streams; mapping the neural state trajectories, behavioral log streams, and environmental pressure vectors to a maritime context-aware latent space with the optimization objectives of maximizing the mutual information of positive sample pairs, minimizing the distance of negative sample pairs, and maximizing the correlation between negative sample pairs and the environmental pressure vector, thereby obtaining a comprehensive contextualized common representation of the crew; and calculating metacognitive alignment loss based on the comprehensive contextualized common representation to adjust the crew's navigation tasks. This method can improve the contextual sensitivity of task adjustment.

The present application provides an asphalt pavement apparent disease evolution deduction method, system, device and medium, which belongs to the field of road engineering, transportation infrastructure operation and maintenance and intelligent detection technology, and comprises the following steps: constructing a unified space-time reference system, aligning and normalizing multi-source data in space-time; forming section-level disease quantification characteristics based on the recognition results of the inspection images and the structural indexes; event coding the maintenance measures of the road surface and pre-processing the section-level disease quantification characteristics corresponding to the time of the maintenance measures; constructing the correlation, time lag effect and spatial consistency between the disease types, disease characteristics and structural indexes to generate the collaborative evolution characteristics of the coupling relationship between diseases; constructing a deep time series prediction model to obtain the disease development trend at a specified time scale in the future, and forming the section-level disease evolution prediction result. The present application can depict the disease development law at different road sections and different time scales, and predict the future disease expansion.

This invention relates to a system for identifying and assessing valvular motion abnormalities in fetal echocardiography, specifically in the field of echocardiography imaging. The system first robustly extracts the valvular central skeleton from a single frame of ultrasound image, overcoming image noise interference. Then, it constructs a physically constrained curve model to simulate the valvular tissue's mechanical properties and precisely tracks its motion trajectory in the image sequence through spatiotemporal optimization. Finally, the system isolates the overall cardiac motion, quantifies the valve's own opening and closing deformation, interleaflet motion synchronicity, and local tissue strain rate, thereby achieving early, objective, and quantitative accurate assessment of fetal valvular motion dysfunction and assisting in clinical diagnosis.

The invention discloses a postpartum comprehensive evaluation method based on multi-source heterogeneous data fusion, solves the problem that modeling and evaluation of postpartum multi-source data cannot be realized in the prior art, and realizes systematic, individualized and prospective comprehensive evaluation of puerpera health. The method comprises the following steps: acquiring initial multi-source heterogeneous data of a puerpera to be evaluated and preprocessing the initial multi-source heterogeneous data; and inputting the data into a pre-trained postpartum multi-dimensional collaborative hypergraph set network model to obtain a comprehensive evaluation result. The model comprises a multi-modal set coding module for extracting each modal feature and converting the modal features into a set family; the cross-modal hypergraph construction module is used for generating hyperedges in combination with priori knowledge and data driving to construct an initial hypergraph; the hypergraph set routing module is used for iteratively updating to obtain nodes and hyperedge set families; the time sequence dynamic evolution module obtains a comprehensive time sequence state set family according to the multi-time-point hyperedge set sequence; and the comprehensive evaluation output module outputs an evaluation result based on the comprehensive time sequence state set family and the hyperedge set family.

The application provides a kind of early warning and positioning method and system for lithium battery internal short circuit fault based on deep learning;Early warning and positioning method includes: step S10, multidimensional data semantic acquisition and working condition adaptive pretreatment;Step S20, cross-scale spatiotemporal feature collaborative extraction and alignment;Step S30, dynamic threshold self-adaptive internal short circuit early warning and feedback optimization;Step S40, topological perception fault accurate positioning and verification;Step S50, embedded collaborative optimization and self-diagnosis deployment: the application can find the early weak signal of internal short circuit fault in time, improve the efficiency and timeliness of battery safety monitoring;Improve the sensitivity of early warning;Enhance the robustness in high temperature, low temperature, fast charging and discharging environment;Can accurately identify the battery monomer and electrode position where fault occurs.

A new electronic lock body, including a lock box and a lock catch assembly arranged on one side of the lock box, the lock box is provided with a lock tongue and a lock tongue transmission assembly for moving the lock tongue, the lock tongue is provided with a magnetic part, the lock catch assembly is provided with a magnetic attraction part, the magnetic part and the magnetic attraction part are attracted to each other, the magnetic attraction part drives the magnetic part in the lock tongue to move and further drives the lock tongue and the lock catch assembly to connect. The design sets up elastic transmission assembly and buckle box, not only realizes the stable connection of lock box and buckle plate, but also guarantees the mute effect in the connection process, avoids the noise interference generated when the traditional electronic lock is closed. The setting of the rotating part further improves the emergency opening ability of the lock, when the electronic opening mode fails, the user can rotate the rotating part and use the mechanical structure to open the lock, so as to ensure the reliability and safety of the lock.

The application provides a SAR image target recognition method based on a saliency map guide, comprising: generating a corresponding target saliency map according to an original SAR image, wherein the target saliency map contains target key information; extracting deep features and shallow features in the original SAR image based on the target saliency map, and guiding the deep features and the shallow features to be fused; refining the fused image features by using a multi-layer hollow convolution to obtain target features; and performing type recognition on the target features by using a full connection layer to obtain a target recognition result. The application avoids noise interference in a background image, improves in-class consistency of target features, can efficiently and robustly realize mining of basic features of a target, and improves precision and efficiency of SAR image target recognition.

The invention discloses an intelligent karst groundwater exploration image processing system. The system comprises an image acquisition module, a geologic body gray level layering enhancement module, a geologic feature stretching module, a karst boundary restoration module, a damage track positioning optimization module, a karst damage track identification model establishment module and a karst groundwater exploration image processing module. The invention belongs to the field of image processing, and particularly relates to an intelligent karst groundwater exploration image processing system, which is characterized in that image gray scale is partitioned, and a attribution degree function is designed to realize directional enhancement guided by geological knowledge; a dynamic hazy rule base is established based on a local window background mean value, and shallow weak anomaly is prevented from being masked; the gray scale stretching slope of the extremely low gray scale region is improved based on the adoption of a dark region enhanced power state spreading function, and weak signals are accurately captured; designing a corner and parietal constraint loss function, accurately matching the irregular contour, and reducing the influence of slight offset of the marking frame; and the karst loss identification accuracy of the karst groundwater exploration image is improved.

This invention belongs to the technical field of 3D modeling for engineering construction. It discloses a method and system for modeling bulldozer operation maps in mining areas based on 5D Gaussian geometry. The method includes: S1 acquiring RGB images and depth data of the earthmoving operation area using an RGB-D camera; applying a contrast enhancement algorithm to the RGB data and vibration compensation to the depth data; and performing semantic recognition of the category of each pixel location; S2 back-projecting the depth data to generate a 3D point cloud; constructing 3D Gaussian units at the sampling points of the point cloud; and converting them into 5D Gaussian implicit units covering spatial, temporal, and semantic dimensions after fusing semantic category and timestamp information; generating a 3D implicit base map; S3 filtering the 5D Gaussian implicit units; and projecting to obtain a 2D rendered image. This invention provides 3D visualization support for bulldozer operation in mining areas, is applicable to scenarios such as mining, and solves modeling problems such as weak texture in earthmoving and equipment vibration interference.

The invention discloses a photovoltaic cell panel defect detection method based on deep learning, and the method comprises the steps: carrying out the synchronous collection and registration of a visible light image and a radio frequency envelope image under pulse current excitation, and forming four-channel input; yOLOv10n is used as a base line to construct an LJH-YOLO lightweight network, and through four improvements of C2fMsFaster EMA, FDPN, CARAFE and Focaler-PIOU, the LJH-YOLO lightweight network is constructed on the basis of the parameter quantity 1t, the parameter quantity 1t, the parameter quantity 1t, the parameter quantity 1t, the parameter quantity 1t and the parameter quantity 1t; the multi-scale defect feature extraction capability is improved under the 3M condition; performing secondary confidence coefficient correction on the reasoning result by using the radio frequency power mean value, and reducing optical pseudo defect over-killing; the pulse current, the radio frequency antenna and the training super parameter all give interval ranges, and can be directly embedded into an existing production line. The implementation result shows that when the method is operated at the edge end of 30 fps, the detection rate of seven types of defects such as cracks and broken gates is superior to that of a base line, and the beat requirement of a gigawatt-level production line is met.

The invention discloses a vehicle mobile phone volume change track detection method, device and equipment and a storage medium. According to the method, a Python automation script is used for controlling terminal equipment to execute a sound detection process, a target sound source is automatically triggered, a volume gear is adjusted, a sound scene is switched, and an operation environment is provided for sound data collection; according to an operation environment, a browser sound collection interface is synchronously called to capture physical parameters of loudspeaker sound of terminal equipment in real time, original collection data are obtained, the original collection data are calibrated and filtered, standardized sound data are generated, and the standardized sound data are displayed in real time through a visual interface. And after the detection is completed, the full-amount detection data is automatically packaged and stored according to the preset format, so that the full-automatic operation of the volume change track detection of the vehicle-mounted terminal and the mobile phone can be realized, the detection efficiency is remarkably improved, the subjective error of manual operation is eliminated, the environmental noise interference is effectively overcome, and the objectivity and accuracy of the detection data are ensured.

This invention proposes a hierarchical label text classification method based on the MOE model, comprising the following steps: encoding the input text using a shared encoder to obtain a unified vector representation of the text; inputting the text vector representation into a router module to calculate the probability distribution of the first-level labels, and activating the top-k expert modules corresponding to the first-level labels; the activated expert modules performing classification predictions within their respective second-level label subspaces; and weighted fusion of the prediction results from the expert modules to obtain the final hierarchical label classification result. This invention extracts text semantic features uniformly through a shared encoder, avoiding redundant learning across branches; it allows samples to simultaneously activate multiple candidate experts through a Top-k soft routing mechanism, breaking the rigid single-path limitation of traditional top-down methods; and it combines routing weights with expert outputs, using the maximization of joint probability as the decision criterion, significantly reducing the number of model parameters and the risk of error cascading while ensuring classification accuracy.

This application relates to a method, apparatus, computer device, and storage medium for determining contextual hot words. The method includes: processing the speech to be recognized and candidate hot words in a candidate word list to determine the frame-level speech vector and the text semantic vector of each candidate hot word at each time step; performing temporal accumulation processing of boundary contributions on the frame-level speech vectors to determine an acoustic boundary sequence, wherein the acoustic boundaries in the acoustic boundary sequence are used to indicate the boundary positions of the frame-level speech vectors in the speech to be recognized; performing local alignment of the candidate hot words and the frame-level speech vectors on the acoustic boundary sequence according to the text lexical length of the candidate hot words to determine the acoustic vector group corresponding to the candidate hot words; and filtering each candidate hot word according to the text semantic vector of each candidate hot word and the frame-level speech vector of the acoustic vector group to obtain the contextual hot words of the speech to be recognized, thereby improving the accuracy of contextual hot word recognition.

This application discloses an incremental training method and a computer-readable storage medium for a dental caries diagnosis model. The method includes: acquiring at least one initial dental caries diagnosis image output by the dental caries diagnosis model, wherein the initial dental caries diagnosis image is an image labeled with dental caries diagnosis annotations obtained by the dental caries diagnosis model from an existing dental caries image; responding to a user's correction operation on the initial annotations in the initial dental caries diagnosis image, collecting the user's correction behavior information, and generating a first dental caries diagnosis image labeled with at least one correction annotation; determining a quality score for each correction annotation based on the correction behavior information; constructing an incremental training dataset based on the first dental caries diagnosis image labeled with target correction annotations of high quality scores, and its corresponding initial dental caries diagnosis image; and incrementally training the dental caries diagnosis model based on the incremental training dataset and the original training dataset of the dental caries diagnosis model.

The invention relates to the technical field of chemical or physical testing, in particular to a system for testing residual monomers in photoresist resin, which comprises the following corresponding steps: determining a target spectrum curve of a target sample, and determining a spectrum characterization characteristic value of the target monomer by using the absorption intensity of the target monomer in the target spectrum curve; determining key samples highly associated with the target monomers in the sample set by using the similarity of the target monomers of any two samples in the sample set in the corresponding target wavenumber range; determining a comprehensive residual characteristic of the target sample by using each spectrum characterization characteristic value in the target spectrum curve, and determining a comprehensive residual intensity index of the target monomer in the corresponding sample set by using the key sample and the comprehensive residual characteristic; and based on the comprehensive residual strength index of the target monomer, determining a content detection result of the target monomer in the to-be-detected photoresist resin. According to the technical scheme of the invention, the detection accuracy for measuring the content of the residual monomers in the photoresist resin is remarkably improved.

The invention provides a stirring device of a shield tunneling machine, which solves the problem that accumulated materials are difficult to scrape off through a stirring blade and accumulate for a long time due to the fact that the stirring blade of the existing stirring machine adopts a fixed-length structure and a gap exists between the stirring blade and the inner wall of a tank body, and can be widely applied to the field of stirring machines. The device specifically comprises a tank body, a stirring shaft is arranged in the tank body, a plurality of fixing pipes are distributed on the surface of the stirring shaft, stirring blades are connected into the fixing pipes, first blades are arranged at the top ends of the stirring blades, second blades are connected to the surfaces of the first blades, the shapes of the second blades are matched with the inner wall of the tank body, and the ends of the second blades can be attached to the inner wall of the tank body; the ends of the second blades protrude out of the ends of the first blades, so that when the ends of the second blades are attached to the inner wall of the tank body, only the second blades make contact with the inner wall of the tank body.

An endoscopic imaging system includes a light source unit, an endoscope, a camera unit, and a display. The light source unit includes an excitation light source and a multi-bandpass filter. The excitation light source outputs at least a first wavelength of first excitation light and a second wavelength of second excitation light. The multi-bandpass filter filters the first and second excitation lights. The endoscope contains a multi-bandstop filter and an image sensor. The multi-bandstop filter filters out the first and second excitation lights from the light returning from the area to be observed. The image sensor obtains first and / or second fluorescence from the filtered light and generates a fluorescence signal based on the first and / or second fluorescence. The camera unit generates a fluorescence image based on the fluorescence signal. The display shows the fluorescence image. This endoscopic imaging system, with its multi-bandpass and multi-bandstop filters, can perform imaging based on different fluorescent dyes.

This invention relates to the field of agricultural information technology, specifically to a system for constructing a tropical agricultural resource database based on agricultural big data. The system includes: a data preprocessing module, a data tagging module, a genealogy management module, a correlation analysis module, and a data verification and supplementation module. By tagging data in multiple dimensions according to tropical crop types, geographical and climatic zones, and production management stages, and mapping the tagged data to corresponding hierarchical nodes in a pre-defined tropical agricultural resource genealogy framework, the system achieves structured reorganization of data according to domain knowledge. Based on this, the system performs cross-data source correlation analysis within the same node of the genealogy framework, establishing precise data correlation links. This solution solves the problems of loose organization of tropical agricultural data and low accuracy of cross-source correlation, effectively improving the structured integration of data resources and the credibility of data analysis results.

The application provides a non-invasive physical examination equipment multi-modal data AI integration method and system, relates to the technical field of data processing, and comprises the following steps: collecting physiological signals and morphological structure data through a non-invasive physical examination equipment, extracting time-frequency domain and morphological characteristics and standardizing to obtain an initial characteristic set, calculating mutual information and orthogonally decomposing to obtain shared characteristic components and unique characteristic components, determining associated nodes to establish hyperedges, nonlinearly fusing weighted aggregation of nodes in the hyperedges to obtain hypergraph embedding characteristics, extracting a time sequence evolution mode matching knowledge base to obtain a candidate diagnosis path, calculating conditional probabilities of each node based on characteristic activation values and performing chain propagation to obtain path joint probabilities, obtaining a posterior probability distribution through variational inference and determining an effective diagnosis path, and normalizing the posterior probability to obtain path weights to solve a health state evaluation result.

This invention discloses a dynamic feedback recommendation method and apparatus based on reinforcement learning. The method includes: pre-training a user behavior simulator using causal inference technology to generate a counterfactual user-item bipartite graph showing complete user-item interaction; weighted aggregation of the real user-item bipartite graph and the counterfactual user-item bipartite graph, using Top-K sampling to generate an interactively balanced user-item bipartite graph; using graph convolution to aggregate neighborhood user information of items to obtain graph-enhanced action representations; using gated recurrent units and self-attention mechanisms to process the interaction sequence between the user and the recommendation system to obtain user state representations at each time step; predicting Q-values ​​based on action representations and user state representations; optimizing the action decisions of the recommendation model based on user reward feedback; and finally achieving dynamic feedback recommendation of Internet information. The apparatus includes a processor and a memory. This invention utilizes causal inference and graph convolution techniques to introduce interactively balanced user-item bipartite graph information, enriching and enhancing agent action representations and user state representations through graph structure association.

The application discloses a kind of based on frozen semantic embedding guide's generative recommendation supervision fine-tuning data construction method.The application avoids the distortion of semantic space and noise interference of collaborative training by freezing the semantic embedding obtained externally as a stable anchor, effectively protecting the content priori knowledge. At the same time, on the basis of maintaining semantic consistency, the collaborative neighborhood and sequence transfer preservation loss are introduced, so that the item representation can still completely retain the behavior patterns and interest evolution law in the original collaborative filtering space during the alignment process to the semantic space, obtaining high-quality fusion representation with semantic enhancement and collaborative authenticity. Based on the fusion representation, it is discretized into multi-level semantic identifiers through residual clustering, and instruction-response samples suitable for large language model supervision fine-tuning are constructed, realizing the effective migration of traditional recommendation knowledge to generative recommendation paradigm, enabling large language model to have sequence recommendation capability, and significantly improving the recommendation effect in cold start scenario.