48results about How to "Avoid dilution" patented technology

The present application belongs to the technical field of internal combustion engine, and particularly relates to a kind of alcohol hydrogen fusion engine precombustion chamber system and its control method.Alcohol hydrogen fusion engine precombustion chamber system includes the engine main body with the main combustion chamber and precombustion chamber being communicated with each other, and also includes methanol supply unit, hydrogen production unit, hydrogen delivery control unit, exhaust gas recirculation unit and main control unit;Main control unit collects engine operating parameters and judges current working condition, and adjusts methanol supply path, hydrogen supply path, precombustion chamber working mode, hydrogen production unit heating mode and on-off state of exhaust gas recirculation unit according to working condition.The present application can realize engine full-condition combustion performance optimization without complex auxiliary device, and also considers cold start reliability, high and low load combustion stability and operation energy efficiency, to improve the whole machine operation reliability and transient response capability.

This invention discloses a structured compression method that guarantees the fidelity of knowledge objects, comprising: writing the original memorized text into an independent read-only storage area; extracting typified knowledge objects and marking their evidence intervals in the original text; recording the precise character intervals of detected behavioral constraints and setting compression lock flags; calculating the irreplaceability score of each knowledge object, routing according to the score and knowledge type, and performing hierarchical differentiated compression; reconstructing the original text based on the read-only storage area and evidence intervals, performing decompression on the compressed knowledge objects, calculating the F1 score of the decompressed text and the original text, performing implication detection on the behavioral constraints, and reverting to the previous version if a preset threshold is not reached; in multi-round cascaded compression, forcibly passing through knowledge objects whose immutable markers are true. The structured compression method of this invention guarantees the complete preservation of behavioral constraints, verifiable fallback of facts, and no distortion of the program step structure.

This invention discloses a method for wildlife identification and behavior analysis monitoring based on cross-modal data fusion, belonging to the field of wildlife monitoring technology. The method includes: synchronously acquiring multi-source heterogeneous data with a unified time reference; outputting alignment features through a deformable spatiotemporal alignment module, with the alignment learning rate dynamically adjusted by downstream uncertainty feedback; extracting multi-scale visual and audio features; adding uncertainty prediction branches to each modality, calculating dynamic fusion weights, and enhancing private features; constructing a dynamic spatiotemporal graph, using uncertainty differences as edge weights, and outputting enhanced node features through graph convolution; matching with a multi-prototype comparison library, with temperature parameters dynamically adjusted by uncertainty, outputting behavioral labels or candidate anomalies; and end-to-end optimization of the joint loss function. This invention, with uncertainty as its core, achieves closed-loop collaborative optimization across the entire link from alignment, fusion, graph interaction to identification, significantly improving the robustness and accuracy of wildlife behavior identification in complex field scenarios.

The present application belongs to the technical field of plastic waste recycling, and discloses a plastic waste recycling device, which comprises a feeding box, a separation frame, a collecting cover, a cleaning cover, a dryer, a fixing frame and a linkage frame. The inner wall of the separation frame is provided with a groove, a separation rod, a guide plate and a sliding shaft. The top is assembled with a reciprocating screw rod, a moving rod, a spring telescopic rod and a magnetic cleaning brush through the fixing frame and the linkage frame, and is driven by a servo motor to realize reciprocating cleaning. The inner wall of the cleaning cover is connected with a pipe at both ends. The inner wall of the pipe close to the dryer is conical. A cleaning corrugated cylinder and an auger are arranged inside. A driving assembly is arranged on the outside. The dryer is connected through a conveying pipe. The present application realizes accurate separation of metal and plastic through the magnetic attraction reciprocating one-way cleaning structure, solves the problem of plastic retention, adopts the conical connecting pipe leak-proof design to reduce the loss of cleaning liquid, cooperates with the corrugated cylinder and the auger for cleaning, greatly improves the cleaning efficiency and the recovery purity, has strong structure linkage, continuous and efficient treatment, and significantly improves the practicality.

The present application relates to the technical field of water body bacteria inactivation, and particularly relates to a bacteria inactivation method based on micro-nano bubble ozone synergistic reaction, comprising the following steps: mixing and conveying ozone gas dissolved water and initial sewage to a front reactor to output as dissolved air sewage; conveying the dissolved air sewage to a reaction tank; obtaining the pressure of a plurality of pressure sampling positions of a conveying pipeline conveying the dissolved air sewage; determining the merging state of micro-nano bubbles in the conveying pipeline; determining an abnormal area in the merging state of the micro-nano bubbles; determining the bubble merging amount of the abnormal area based on the obtained conveying pipeline pressure of the position where the connecting port is located and the average pressure in the conveying pipeline behind the connecting port; determining the dosing amount of alkaline substances and the replenishment amount of the dissolved air water; and continuously conveying the dissolved air sewage from the front reactor to the reaction tank to complete the bacteria inactivation operation, so that the cavitation effect of ozone micro-nano bubbles is strengthened and the solubility of ozone in water is improved.

The application discloses a method for determining the content of combustible gas in steam. The sample is collected by the drainage gas collection method: the gas collection container filled with distilled water is inverted in the sink, steam is introduced to condense, the original water in the bottle is replaced and discharged by using the condensed water, so that the non-condensable combustible gas is collected at the top of the bottle, and the air is isolated by the water seal. Then, the standardized syringe sampling is carried out: after being cleaned and lubricated for multiple times, the airtight syringe is used to pass through the bottle plug and draw a certain amount of gas under the liquid surface, and is quickly sealed to avoid pollution. Finally, the sample is analyzed by the gas chromatograph, and the content of the combustible gas in the steam is quantitatively calculated based on the calibration curve or the correction factor established based on the known concentration standard sample. The application has the advantages of strong anti-interference, standard operation, good sample representativeness and low cost, and is especially suitable for accurate monitoring of low-concentration combustible gas.

The utility model relates to the technical field of biological pest control, specifically is a kind of pest luring-infection device, attract pest into device and automatically infect fungus, specific structure is: the lower end of baffle assembly is equipped with conical tube assembly from top to bottom, connection mechanism is sequentially equipped between conical tube assembly, the inside of conical tube assembly is corrugated and is set, is lure part;Conical tube assembly lower end is connected with tubular member and corrugated pipe, and tubular member and corrugated pipe are infection part, and the inside surface of infection part is covered with the medicament containing pathogenic fungus spore.The application can complete the automatic inoculation of pathogenic fungus after pest adult enters the device, and allow adult to escape after infection, return to population and further horizontally spread fungus spore, to reduce the number of pest population, reach the purpose of prevention and control.The device is simple to operate, low in cost, reusable, and has very broad application prospect.

This invention discloses a carbon-based fuel gasification method based on the principle of a galvanic cell, comprising the following steps: (1) forming a three-layer self-layered structure of a metal layer, a metal oxide layer, and a molten carbonate layer from bottom to top in a reactor; (2) adding carbon-based fuel to the molten carbonate layer; (3) gasification step: forming a galvanic cell with carbon-based fuel as the negative electrode, metal oxide as the positive electrode, and molten carbonate as the electrolyte, wherein the oxygen in the metal oxide is converted into oxygen. 2‑ The form diffuses through the molten carbonate layer to the carbon-based fuel interface, selectively oxidizing the carbon-based fuel to generate CO. The metal oxide is reduced to metal and sinks to the metal layer; (4) Regeneration step: a gaseous oxidant is introduced into the metal layer to oxidize the metal to metal oxide, replenishing the oxygen storage capacity of the metal oxide layer, and completing the storage of oxygen and the circulation of the reaction medium. This invention utilizes the potential of the galvanic cell to spontaneously drive the reaction and manages the gasification process by controlling the "charging / discharging" cycle. It has the advantages of high CO selectivity, good metal oxide volatilization inhibition effect, high fuel impurity tolerance, high gas production calorific value, and spontaneous and efficient reaction, which is suitable for the needs of clean and efficient conversion of distributed carbon-based fuels.

The present application relates to the technical field of metal product finishing, and discloses a metal vacuum cup burr-free ring cutting device with smooth cutouts, which comprises a base, a first support frame fixedly connected to the top of the base, a second support frame fixedly connected to the side of the base close to the first support frame, a ring-shaped plate fixedly connected to the end of the first support frame away from the base, a ring-shaped electric sliding rail fixedly connected to the side of the ring-shaped plate, a first sliding plate and a second sliding plate slidably connected to the inner part of the ring-shaped electric sliding rail through electric sliding blocks respectively, a laser cutting assembly fixedly connected to the side of the first sliding plate, and a connecting plate fixedly connected to the side of the first support frame; and a supporting device fixedly connected to the side of the ring-shaped plate. The metal vacuum cup burr-free ring cutting device with smooth cutouts can realize cooling of the cutting area and avoid blowing cooling gas directly to the laser cutting area.

The application relates to the technical field of edge computing and multi-modal data processing, and discloses a park multi-modal data stream alignment and caching method based on event driving, which improves alignment accuracy, resource efficiency and real-time performance, can realize semantic understanding and association, and meets asynchronous data fusion, and comprises the following steps: S1, detecting an event and generating an event anchor point; S2, based on the event anchor point, dynamically intercepting a segment related to the event in a multi-modal data stream and aligning on a time axis; S3, packaging the aligned multi-modal data segment into an event data packet, caching the event data packet according to an event priority strategy, and querying on demand. The application also discloses a system for implementing the above-mentioned park multi-modal data stream alignment and caching method based on event driving.

The present application relates to a method for synthesizing tinidazole deuterium analogues, such as shown in formula I, wherein one or more of R1-R5 are deuterium and the rest are hydrogen, comprising the step of preparing the tinidazole deuterium analogues from intermediate B by oxidation with meta-chloroperoxybenzoic acid, wherein the oxidation is carried out under low temperature and alkaline conditions, and the intermediate B is shown in formula II: The method for synthesizing tinidazole deuterium analogues according to the present application can avoid the problems of deuterium loss and isotopic dilution during the reaction process, obtain stable isotope-labeled tinidazole deuterium analogues with an isotopic abundance of >98%, and have a high yield. The prepared tinidazole deuterium analogues can be used as isotopic internal standard reagents for quantitative detection of tinidazole, for example, in the isotopic internal standard method for detecting the residual amount of tinidazole in meat products, and have the advantages of high recovery efficiency from the matrix and green safety.

The utility model discloses a kind of device for reducing helium waste, belong to helium leakage test technical field, including cavity, pipeline system, recovery tank, recovery pump, compressed storage mechanism, first support and second support, second support is set to first support right side, cavity is installed in second support, pipeline system gas inlet end is connected cavity, recovery tank is installed on first support, it is provided with three gas inlets, one of which is connected with the gas outlet of one pipeline system, recovery pump is installed in first support, recovery pump gas inlet end is connected with the gas outlet of another pipeline system, pump gas outlet is connected with the second gas inlet of recovery tank through pipeline system, compressed storage mechanism is set in first support, compressed storage mechanism gas inlet end is connected with the gas outlet of recovery tank through pipeline system, compressed storage mechanism gas outlet is connected with cavity through pipeline system. The device can maintain the micro-positive pressure inside system, prevent air backflow, avoid helium concentration dilution, realize reducing helium consumption.

The utility model relates to the technical field of rubber tube processing, in particular to a spraying device for processing a separant in a rubber tube, and adopts the technical scheme that an air drying mechanism is arranged between a cleaning water tank and a separant spraying tank; and the air drying mechanism comprises an annular seat through which the rubber pipe can conveniently pass, and an air pump for supplying air into the annular seat through an air supply pipe, so that the problem that water on the surface of the rubber pipe is easy to dilute the separant concentration when the separant is sprayed on the rubber pipe is solved. The air drying mechanism is arranged between the cleaning water tank and the separant spraying tank, and a water film on the surface of the rubber tube is blow-dried by the air drying mechanism when entering the separant spraying tank, so that the separant is prevented from being diluted, and the subsequent vulcanization quality of the rubber tube is ensured.

The present application belongs to the technical field of wastewater sampling, in particular to a water body sampling device for industrial wastewater tracing, comprising a mounting part, a sampling mechanism, a detection mechanism, a wiping mechanism and a backflow mechanism; the mounting part comprises a buoyancy disc and a mounting box, the buoyancy disc is fixedly installed on the upper end face of the mounting box; the sampling mechanism comprises a control pump, a conveying pipe and a filter, the control pump is fixedly installed in the buoyancy disc, the filter is connected with the input end of the control pump, and the conveying pipe is connected with the output end of the control pump; the detection mechanism comprises a carrier and a sensing plate, the sensing plate is obliquely installed on the outer wall of the carrier, and the carrier is fixedly installed on the inner wall of the buoyancy disc; the wiping mechanism comprises a sliding block and a wiping layer, the sliding block is slidingly arranged on the outer wall of the carrier, and the wiping layer is elastically arranged on the bottom of the sliding block; the backflow mechanism comprises a storage cylinder and a backflow pipe; through the cooperation of the above structure, the accuracy and reliability of the industrial wastewater tracing data are improved, and the online continuous sampling and detection mode truly meets the high-precision requirement.

The utility model relates to the technical field of termite trapping equipment, in particular to a termite trapping and killing integrated device which comprises an outer cylinder, a trapping cavity used for collecting termites is formed in the outer cylinder, a cover plate is installed on the top of the outer cylinder, and a plurality of trapping openings allowing the termites to enter are formed in the outer side of the cover plate. The inner wall of the trapping cavity is provided with a plurality of ultraviolet lamps capable of killing termites, the ultraviolet lamps are externally connected with a control switch through wires, the control switch controls the ultraviolet lamps to be turned on and turned off, and the inner side of the bottom of the outer cylinder is provided with a termite trapping agent used for trapping the termites. According to the termite trapping and killing integrated device, through multiple trapping design, the attraction effect on termites is remarkably improved. The termite trapping agent arranged on the inner side of the bottom of the outer cylinder can emit specific smell and generate strong attraction to termites; and a sweet coating is arranged on a pine plate of the bait assembly in the middle of the trapping cavity, so that the attraction to the termites fond of sweet food is further enhanced.

The utility model discloses a shower device, and belongs to the technical field of shower equipment. The shower device comprises a pipeline structure, a switching valve and a foam generator, the pipeline structure comprises a water inlet path, a first water passing path, a second water passing path and a mixed water outlet path, and the switching valve is used for enabling the water inlet path to be communicated with the first water passing path and the second water passing path or enabling the water inlet path to be communicated with the first water passing path. A foaming assembly of the foam generator is installed between the first water passing path and the mixed water outlet path. By adjusting the position of the switching valve, the shower device can be switched between the clean water outlet mode and the foam liquid outlet mode, a user does not need to smear and rub the foam liquid by himself, and the cleaning efficiency and the use experience of the user can be improved. And moreover, the first water passing path and the second water passing path in the clean water outlet mode are both in a conducted state, and water flow in the second water passing path can clean residual foam liquid in the foaming assembly, so that residual cleaning products in the foaming assembly and the mixed water outlet path are avoided.

The application relates to the field of data processing and discloses a wire surface defect judgment method based on object recognition, which comprises the following steps: determining a wire center axis track and establishing a servo coordinate system to convert an image; adopting an anisotropic derivative operator filter to compress in a dimension perpendicular to an axis direction, determining topological continuous nodes or discrete noise points according to the consistency of adjacent row gradient polarity, respectively implementing self-gain accumulation or zero assignment truncation to generate a projection sequence; determining a dynamic threshold according to the projection sequence and marking an interval, inversely mapping to extract a local image block to judge a defect category, and the application realizes nonlinear accumulation of defect feature energy through feature pre-screening and feedback closed-loop regulation, avoids noise dilution of weak contrast defects, reduces a missed detection rate, and improves system real-time processing throughput.

This application provides a decision-making method, apparatus, electronic device, medium, and program product that can be applied to the field of artificial intelligence technology. The method includes: acquiring a decision-making matter; performing multi-dimensional analysis on the decision-making matter to obtain corresponding confidence scores; determining a first support quantity for approving the decision-making matter and / or a second support quantity for disapproving the decision-making matter based on the confidence scores, and calculating a current consensus value based on the first support quantity and / or the second support quantity; acquiring at least one environmental variable to characterize the external environment state, and determining an environment-adaptive consensus threshold based on the environmental variable; performing target risk detection on the decision-making matter to obtain target risk detection results; and generating a final decision for the decision-making matter based on the current consensus value, the environment-adaptive consensus threshold, and the target risk detection results.

The application discloses a commodity implicit association strength calculation method based on multi-scale feature fusion, belongs to the technical field of commodity data analysis, and aims to solve the problems of single feature utilization, rough multi-scale feature fusion and insufficient capture of commodity implicit association in the prior art; the application carries out preprocessing on collected commodity multi-dimensional original data, extracts features of three scales of commodities, i.e., fine, medium and coarse, and converts the features into preset dimension feature vectors, after feature optimization, completes deep feature fusion through a hierarchical mode of same-scale fusion-cross-scale attention weighted fusion, then calculates the association strength between commodities and determines the association type, finally outputs the result and completes algorithm closed-loop optimization; the application significantly improves the calculation precision and recall rate of commodity implicit association, takes into account the result explainability and real-time performance, and is suitable for the application requirements of multiple scenes of e-commerce.

The application discloses an intelligent closestool and a descaling method thereof, and comprises the following steps: S1, starting descaling; S2, switching a reversing valve downstream of an instant heater to a state of opening a cleaning port communicated with a water spraying body; S3, starting a liquid pump connected with a switching valve upstream of the instant heater, and the liquid pump stays for T1 time in the state of opening the cleaning port of the reversing valve, so as to pump descaling liquid in a liquid storage tank communicated with the liquid pump to the water spraying body and discharge the descaling liquid from a water outlet of the water spraying body; S4, switching the reversing valve to a state of opening a descaling port communicated with the liquid storage tank, and the liquid pump is in an open state for T2 time, so that the descaling liquid circulates in a loop formed by the liquid storage tank, the liquid pump, the switching valve, the instant heater and the reversing valve for T2 time. By arranging steps S2 and S3, the descaling liquid in the liquid storage tank can be used to replace residual cleaning water in a passage between the switching valve and the water spraying body, so that dilution of the descaling liquid by the residual cleaning water in the water passage is avoided, and the descaling effect is effectively improved.

The utility model discloses a steam condensate water separation structure for rice wine distillation, and relates to the technical field of rice wine distillation. The device comprises a cooling shell, a vapor-liquid separation assembly and a water temperature control assembly, a condensation pipe is arranged in an inner cavity of the cooling shell, and the top of the condensation pipe penetrates to the outer side of the cooling shell. Through the vapor-liquid separation assembly, the sealing ring and the plugging ring which are arranged in the separation shell are matched with the sealing plug and the plugging block which are driven by the air cylinder, stage separation is achieved in the vapor condensation process, and when the sealing plug is lifted, the plugging block seals the central through hole of the plugging ring; when the air cylinder is reset, the sealing block releases the sealing of the through hole, the wine liquid is discharged in a one-way mode through the liquid discharging pipe, and the residual steam is independently pumped away through the exhaust mechanism. The steam can be effectively prevented from being mixed with the wine liquid after being secondarily condensed, and the wine body is prevented from being diluted.

The invention provides an EEG signal feature extraction method and system, a storage medium and a terminal. The method comprises the following steps: preprocessing an EEG signal; constructing a multi-scale reference window and a multi-scale test window, and calculating a multi-scale spectrum structure change score of the preprocessed EEG signal; detecting candidate segment boundary points of the preprocessed EEG signal based on the multi-scale spectrum structure change score; on the basis of the candidate segmentation boundary points, EEG adaptive segmentation is obtained; performing Hermite function decomposition on the EEG signal in the EEG adaptive segment to obtain EEG segment level features; and constructing an EEG feature vector based on the EEG segment-level features and the EEG global statistics. According to the EEG signal feature extraction method and system, the storage medium and the terminal, EEG signal features with the characteristics of high event sensitivity, high adaptive ability, excellent stability and the like can be extracted.

This invention provides a chemical detection system for scraping residue on printed tape, relating to the field of data processing technology. The system includes: identifying residue categories in a sample; ranking each residue category by risk to obtain a risk ranking result; correcting the risk ranking result using spatial anisotropic weighting to obtain a key residue selection list; and a judgment module, used to establish quantitative detection models for each residue category based on the key residue selection list, analyze the sample, and obtain the measured content value of each residue category; correcting each measured content value using spatial anisotropic weighting to obtain a corrected content value; and comparing the corrected content value with the safety threshold of the corresponding residue category to obtain a judgment result. This invention achieves effective identification, accurate quantification, and reliable judgment of scraping residue on printed tape even when the distribution is uneven.

The utility model discloses a kind of automobile exhaust detection devices, including tail gas analyzer, the one end of tail gas analyzer is connected with conveying pipe, the other end is connected with discharge pipe, the inside of conveying pipe is slidably installed with telescopic pipe, the end of telescopic pipe is installed with detection probe, sealing clamping mechanism is equipped between detection probe and telescopic pipe, and lifting transmission mechanism is installed below tail gas analyzer.The utility model has following effect: when detecting exhaust, detection probe is inserted into automobile exhaust pipe, through sealing clamping mechanism, the automobile exhaust pipe of different diameter size can be firmly clamped, and the gap between exhaust pipe and detection probe is tightly sealed, prevent outside air from mixing when exhaust detection, significantly improve the sealing and reliability of exhaust sampling process, improve overall detection precision.

The application discloses a control method and system of a multi-in-one intelligent distributed power supply acquisition safety terminal, relates to the technical field of intelligent control of power distribution networks, and converts the topology structure of a power distribution network line into a graph structure to obtain a topology graph structure; key time nodes in historical operation data are identified, time sequence attention weights are calculated to weight the historical voltage sequence, and weighted historical data are obtained; the topology graph structure and the weighted historical data are input into a graph neural network for feature aggregation to obtain node embedding representation; a target line is identified according to a comprehensive risk value, a multi-objective optimization function is constructed based on the node embedding representation for solving, and an initial access capacity scheme is obtained; the initial access capacity scheme is substituted into a digital twin model for simulation, an anti-risk coefficient is calculated, and the anti-risk coefficient is corrected in combination with a load threshold value and a reverse power threshold value to obtain an optimal access capacity scheme. The accuracy of distributed power supply access capacity evaluation and the optimization level of the carrying capacity of a power distribution system are significantly improved.

The invention relates to an intelligent event analysis method and system based on AI, and the method comprises the following steps: obtaining multi-modal original data related to an event, and extracting the feature representation of each modal data; the method comprises the following steps of: extracting features of an event, constructing a hierarchical multi-modal feature graph containing intra-modal sub-graphs and a cross-modal association graph on the basis of the extracted features, sequentially performing intra-modal GAT aggregation and modal perception cross-modal attention calculation by utilizing a graph attention network to obtain global features after cross-modal aggregation, and finally outputting event type probability distribution through a classification layer and identifying event types; and based on the identified event type and the multi-modal feature graph, constructing a layered multi-modal event knowledge graph, analyzing a logical relationship and an evolution path between the events by using time sequence modeling and a causal reasoning model, and outputting an analysis result. Compared with the prior art, the method has the advantages that the multi-modal event type identification precision and accuracy can be improved, the analysis efficiency is improved, and the like.

The utility model relates to infant hygiene product technical field, especially infantile urine sample taking paper diaper, including paper diaper body, the middle of paper diaper body front is provided with the absorption core layer, the both sides of paper diaper body are provided with the leakage prevention and isolation edge, the middle of paper diaper body back is provided with the paste ring, the paste ring is arranged in the urine position of infant. Through setting the material taking hole in the absorption core layer, and setting the material taking film under it, form independent sampling chamber, urine sample flows into the material taking film in the material taking hole and is saved separately, can effectively collect the original urine sample under the premise of not interfering with the normal urination process of infant, avoids being absorbed material adsorption, dilution or pollution, ensures the integrity and the cleanliness of sample, is favorable to the accuracy of subsequent detection.

The application discloses an automatic driving training method based on a difficult case pool and a decision model, and belongs to the technical field of driving training. The application comprises the following steps: S1, obtaining automatic driving observation information, training, obtaining driving actions, and calculating trajectory-level evaluation indexes; S2, determining a difficult case pool according to the trajectory-level evaluation indexes; S3, generating a difficult case robust training branch according to the difficult case pool; S4, updating an automatic driving decision model according to a regularization term of the difficult case robust training branch and a training loss of a main training branch, and obtaining a robust automatic driving decision model; and S5, outputting driving actions or behavior decisions by using the robust automatic driving decision model. The application comprises a closed-loop training mechanism composed of ordinary sample training, difficult case identification, difficult case pool updating, difficult case sampling, strong disturbance triggering and opponent robust regularization, and can simultaneously consider normal driving capability and robustness under observation disturbance.

This application discloses a radar semantic segmentation method and system based on explicit Doppler coding and physical constraints, relating to the fields of intelligent sensing and radar signal processing technology. The method includes: acquiring raw range-angle-Doppler spectrum data from a millimeter-wave radar and preprocessing it; extracting Doppler motion features through baseline estimation and spectrum compensation, and generating a motion-aware feature map and a dynamic saliency weight map; fusing the data through self-attention and cross-attention mechanisms to generate a multi-view fusion feature map containing motion semantic saliency; wherein, for the multi-view fusion feature map at continuous time points, a dynamic persistence constraint loss function is constructed to apply time consistency regularization to the Doppler motion features. The multi-view fusion feature map is input into a semantic decoding network, and the radar semantic segmentation result is output. This application effectively improves the accuracy, stability, and physical consistency of radar semantic segmentation in complex dynamic scenes.

The application discloses a general performance enhancement distillation method for a convolutional neural network, and relates to the technical field of artificial intelligence.The method first acquires feature maps output by teacher-student models, generates an importance map through channel projection and activation, and calculates two types of complementary spatial gains, namely consistency and coverage; a learnable pixel-level gate consensus module is used to model the correlation between teacher and student features, and a mixed coefficient is adaptively output; a dynamic spatial consensus weight mask is generated through convex combination, a feature distillation loss term is constructed in combination with the mask, a gating regularization term is introduced to construct a total loss function, and a student model is iteratively optimized.The application solves the problems of rigid weighting rules, insufficient adaptation in the training stage, and inaccurate gradient distribution of existing methods, realizes adaptive matching of the distillation strategy and the training stage, significantly improves the precision and generalization ability of a lightweight student model in a dense target detection task without increasing inference overhead, and is suitable for resource-constrained platforms such as edge devices.