25results about How to "Addressing slow performance" patented technology

The application discloses a visual language navigation method based on adaptive prompt information during testing, comprising the following steps: obtaining natural language-based navigation instructions and 360-degree panoramic visual observation information of an intelligent agent; extracting low-frequency visual features of sub-images and selecting valuable information from an image memory bank for prompting; extracting a text feature vector corresponding to the navigation instructions and an initial state vector of the current environment; selecting historical information from the visual feature memory bank using the state vector for prompting; inputting the state vector and the instruction features into a visual-language interaction module to generate navigation action selection; updating the normalization layer in the full connection layer according to the prediction score of the navigation action selection; and repeating the above steps until the intelligent agent stops selecting or reaches the maximum number of movement steps. The application can effectively enhance the navigation performance of the pre-trained intelligent agent model in unknown test environments.

The invention discloses low-carbon polypropylene fiber concrete with high toughness and crack resistance, and belongs to the field of concrete. The raw materials comprise, by mass, 100-110 parts of coarse aggregate, 60-70 parts of fine aggregate, 20-30 parts of cement, 80-120 parts of fly ash, 15-20 parts of modified polypropylene fiber, and 2-3 parts of a water reducing agent; the use amount of the water meets the condition that the water-cement ratio is 0.3-0.4; according to the method, the surface of the polypropylene fiber is subjected to controllable micro-etching through concentrated phosphoric acid, a microscopic gully and micropore structure is formed, effective roughening of the surface of the fiber is achieved, the specific surface area of the fiber is greatly increased, the fiber can generate a stronger mechanical meshing effect in concrete, pulling out of the fiber is effectively prevented, and the tensile strength of the fiber is improved. And in the subsequent silane coupling agent treatment, a stable'molecular bridge 'is constructed between the fiber surface subjected to acid etching activation and an inorganic cement matrix, and firm Si-O-Si covalent bond connection is formed, so that the problem of weak bonding force between the common polypropylene fiber and the matrix interface is solved.

The invention discloses a multi-mode signal real-time self-adaptive recognition and extraction method for a complex scene, solves the problem that an existing interference suppression and signal extraction algorithm cannot adapt to gravitational wave measurement requirements, belongs to the technical field of digital signal processing, and adopts a fixed sequence of main carrier clock sideband pseudo code signals to perform signal extraction and separation. According to the sequence, the processing efficiency is optimized, the weak signal is prevented from being covered by the strong signal, and the interference of subsequent processing is remarkably reduced. Compared with parallel or disordered processing in the background technology, the method provided by the invention has the advantages that strong interference components are stripped in sequence, so that rapid convergence in real-time operation is ensured, and the robustness and precision of overall signal extraction are improved.

The application discloses a domain self-adaption image classification method based on self-training and centroid alignment, comprising the following steps: obtaining data sets of a source domain and a target domain and performing simple data preprocessing; through a self-training process, the network obtains simple classification capability; through a consistency regularization loss, the consistency of outputs of two classification modules is improved; the distance between target domain data and corresponding class centroids of the source domain is shortened, and the target domain data features are promoted to move to the direction of the source domain class centroids, so that the features are aligned at a structure level; and in the learning process, the classification model effectively aligns the knowledge of different domains gradually. The application combines feature information and structure information, considers two angles of distance and direction, guides the model to learn domain-invariant features, and effectively solves the problem of performance decline of an image classification network caused by large differences between data distribution of the source domain and the target domain.

The embodiment of the application provides a reference signal configuration method, a storage medium, an electronic device and a computer program product, and the method comprises the following steps: a first node determines a target modulation mode of a reference signal in multiple reference signal modulation modes. According to the application, the modulation mode of the reference signal is no longer fixedly configured, and the communication node can select a suitable modulation mode from the multiple reference signal modulation modes according to the needs, so that different communication scenes and needs can be more flexibly adapted, and the problem that, in the related art, the nonlinear difference between the reference signal and the data signal fixedly using low-order modulation in the scene of high-power transmission and high-order data modulation can cause the performance of the receiver to decline when the receiver performs channel estimation and data recovery is solved, and therefore the effect of improving the efficiency and performance of the communication system is achieved.

This invention relates to the fields of unmanned aerial vehicles (UAVs) and computer vision technology, and particularly to a UAV-based method for classifying bridge components. The technical problem is that existing point cloud-based automatic classification methods for bridge components suffer from issues such as scarce samples, high annotation costs, insufficient model generalization ability, and low processing efficiency. The technical solution is a UAV-based method for classifying bridge components, comprising a data acquisition step, a data augmentation step, a data preprocessing step, a model training step, and a component classification step. In the model training step, preprocessed 3D point cloud data and its corresponding component category labels are used to train a deep learning-based point cloud semantic segmentation model, enabling the model to learn the feature representations of different bridge components. This invention uses a systematic data augmentation strategy to simulate real-world acquisition interference, combined with an efficient point cloud deep learning model, to achieve accurate, robust, and efficient classification of various bridge components.

This invention provides a cross-modal anomaly detection and localization method and system based on context and dual text. The method includes: extracting global semantic visual features, spatial fine-grained visual features, and fragment-level visual context information from the video frame to be detected; generating instance-level conditional cue vectors using the fragment-level visual context information as conditions; mapping and generating a first text prototype feature representing a normal state and a second text prototype feature representing an abnormal state; performing cross-modal comparison between the generated dual text prototype features and the extracted global semantic visual features and spatial fine-grained visual features, generating frame-level anomaly scores in parallel to complete anomaly detection, and generating pixel-level anomaly localization heatmaps to complete anomaly localization. This invention enables the generated normal and abnormal text prototype features to adaptively adjust with the context of the current video scene, overcoming the shortcomings of static text cue's lack of scene adaptability and susceptibility to environmental interference.

This invention relates to a wide common-mode sampling circuit, specifically integrated circuit technology. The circuit comprises: a first switch positioned between the non-inverting input terminal and a non-inverting reference point; a second switch positioned between the inverting input terminal and an inverting reference point; a third switch positioned between the non-inverting reference point and an inverting reference point; a first capacitor positioned between the non-inverting reference point and a non-inverting output terminal; a second capacitor positioned between the non-inverting reference point and an inverting output terminal; a third capacitor positioned between the inverting reference point and an inverting output terminal; a fourth capacitor positioned between the inverting reference point and a non-inverting output terminal; a fourth switch positioned between the non-inverting output terminal and a common-mode voltage input point; a fifth switch positioned between the inverting output terminal and the common-mode voltage input point; an equivalent capacitance of the non-inverting capacitor array positioned between the non-inverting output terminal and a ground connection terminal; and an equivalent capacitance of the inverting capacitor array positioned between the inverting output terminal and a ground connection terminal. This invention provides a wide input common-mode range.

The invention relates to the technical field of machine learning, in particular to a time series data analysis system and method based on deep learning, and the system comprises an input preprocessing module which is used for carrying out the standardization processing and segmentation processing of input time series data; the sliding window recurrent neural network module is used for processing the input time sequence data in each sliding window of eight thousand time steps to generate a short-term time sequence feature vector; the dynamic updating recurrent neural network module is used for processing long-term time sequence data; the meta-learning optimization module is used for dynamically updating initialization parameters of the recurrent neural network module through outer loop optimization; and the output fusion module is used for fusing the feature vectors and generating a final time sequence analysis result. According to the method, a weight updating mechanism which only updates the last quarter layer of the recurrent neural network is innovatively designed, the calculation complexity is remarkably reduced, and the contradiction between the calculation efficiency and the model adaptability in long-sequence time sequence analysis is effectively solved while high precision is kept.

The application discloses a rice leaf disease identification method in an open scene, and relates to the technical field of computer vision, and specifically comprises the following steps: step 1: constructing an image enhancement set based on an adversarial attention mask; step 2: defining an entropy toughness index, and constructing a dynamic high-reliability sample library according to the index; step 3: realizing steady-state linear transformation through dynamic weight distribution; and step 4: combining the optimized pre-training classifier and the linear transformation matrix to form a final rice leaf disease identification model. Through the combination of the adversarial mask, the dynamic high-reliability sample library and the steady-state linear transformation, the application effectively solves the problems of poor robustness and difficult field adaptation of the model in an open farmland environment, and improves the performance of the rice leaf disease identification model in an open scene.

The utility model relates to laser cladding nozzle technical field, specifically disclose a height adjustable laser cladding nozzle, the height adjustable laser cladding nozzle at, it includes: nozzle body, hollow air cylinder, the hollow air cylinder sets up in the nozzle body top and is connected with the nozzle body, the hollow air cylinder's axis place is provided with the hollow passage that penetrates the hollow air cylinder, the axis of the hollow passage coincides with the axis of the nozzle body. The utility model has through hollow air cylinder and laser cladding nozzle coaxial arrangement and realizes nozzle vertical adjustment, has obviously promoted the adaptability, the adjustment accuracy and the efficiency of laser cladding nozzle to complex processing surface, has guaranteed laser and the coaxial coupling effect of powder simultaneously, has improved the homogeneity and the quality of laser cladding layer's advantage.

The utility model relates to a heat dissipation structure technical field discloses a kind of metal matrix composite multilayer heat dissipation structure, including multilayer component, the multilayer component includes heat dissipation fin layer, the heat dissipation fin layer bottom is fixedly connected with intermediate heat conduction layer, the intermediate heat conduction layer is fixedly connected between two second connecting pieces, the intermediate heat conduction layer bottom is fixedly connected with heat source contact layer, the heat dissipation fin layer is the metal matrix composite of foamed copper material quality, the intermediate heat conduction layer is the metal matrix composite of carbon fiber material quality, the heat source contact layer is the metal matrix composite of diamond material quality, the heat dissipation fin layer bottom is provided with two fixed components. In the utility model, heat dissipation fin layer passes through increasing heat dissipation area, utilizes air convection to accelerate heat emission, intermediate heat conduction layer can quickly conduct the heat that heat source contact layer transmits to all around, further improves heat dissipation speed, and heat source contact layer is directly contacted with heat source.

This invention discloses an intelligent control method and system for motorcycle calipers. The method involves real-time acquisition and preprocessing of multi-modal data, including vehicle speed, wheel speed, piston displacement, brake disc temperature, tilt angle, and lateral acceleration, using multiple sensors. Based on the data, braking demand and adhesion coefficient are calculated, and tilt angle and lateral acceleration are used to identify straight-line, curved, or emergency braking scenarios. The data is input into a pre-trained braking control machine learning model to generate a target braking force variation curve. Based on this curve, the drive motor and caliper are controlled for braking, and the model is periodically optimized based on braking feedback and historical data. A component degradation model is used to predict the braking efficiency decay trend, and the braking force is compensated and the model is updated. Closed-loop control is used to adjust the motor current in real-time to follow the target braking force. The brake disc temperature is continuously monitored, and thermal fade compensation is triggered when the temperature exceeds a threshold. This invention achieves adaptive, precise, and reliable intelligent braking for motorcycles.

The application discloses a T-shaped gate preparation method based on electron beam exposure, which comprises the following steps: covering two layers of photoresists with different sensitivities on a substrate, exposing and developing to obtain a gate leg photoetching mask structure; depositing metal on the substrate with the gate leg photoetching mask structure; stripping the gate leg metal structure formed by the two layers of photoresist masks to obtain the gate leg attached to the substrate; spin-coating three layers of photoresists on the substrate with the gate leg; forming a gate cap photoetching mask structure for forming a gate cap through the second upper layer of photoresist and the middle layer of photoresist, and forming a gate cap support layer through the second lower layer of photoresist; depositing metal on the substrate with the gate cap photoetching mask and the gate cap support layer mask to obtain a gate cap metal structure; and stripping the gate cap metal structure to obtain a complete T-shaped gate attached to the substrate. The application reduces process damage, avoids the problem that the foot cap dose influences each other when the T-shaped gate is prepared, and is beneficial to the preparation of a T-shaped gate with a shorter gate length.

The invention is applicable to the technical field of metal materials and provides corrosion-resistant enhanced duplex stainless steel and a heat treatment process thereof, and the corrosion-resistant enhanced duplex stainless steel comprises the following raw materials in parts by weight: 90-95 parts of a duplex stainless steel matrix, 3-5 parts of nano titanium nitride, 1-2 parts of modified graphene, 0.5-1 part of niobium powder, 0.3-0.8 part of vanadium powder and 0.2-0.5 part of yttrium powder. According to the preparation method, component optimization is carried out on a duplex stainless steel matrix, treated nano titanium nitride and graphene are introduced, alloy elements such as niobium, vanadium and yttrium are cooperated, and specific solid solution and aging heat treatment processes are matched, so that the corrosion resistance, the mechanical property and the structure stability of the duplex stainless steel are remarkably improved; the problems that traditional duplex stainless steel is prone to precipitation of harmful phases in a high-temperature or corrosive environment, and the performance is reduced are solved.

The utility model discloses a kind of glue filling machine of facilitating colloid individual storage mixing, including storage mixing main part, first support seat, fixed support base, second support seat, rotating gear disc, first motor, second motor and second support seat, the first support seat and the second support seat are fixedly installed on the fixed support base, the rotating gear disc symmetrical rotation meshing connection is arranged in the upper end inside of the first support seat.This utility model two colloid independent storage anti-deterioration system independent sealing storage, by separately arranged first storage barrel and second storage barrel respectively store A / B colloid, sealing cover is combined bolted, completely isolate two-component colloid contact, solve the problem of performance decline caused by mixed glue solution due to chemical reaction, dynamic anti-solidification design, first motor drive meshing rotating gear disc, drive glue storage barrel periodic swing, effectively prevent colloid standing sedimentation or local solidification, significantly extend storage stability.

The application discloses a kind of air supply equipment and gas water heater, it is related to water heater technical field.The air supply equipment includes volute, air supply wheel, motor and heat insulating material piece;Volute is equipped with fan flow channel;Air supply wheel is located in fan flow channel;Motor includes the motor body and motor shaft connected, motor body is located in the side of volute, motor shaft is inserted into fan flow channel, and is transmissionally connected with air supply wheel;Heat insulating material piece is located between motor body and volute, heat insulating material piece is configured to block the heat transfer in fan flow channel to motor body.The technical scheme of the application can prevent motor temperature rise too high, and solve the problem of motor performance decline simultaneously.

The utility model discloses a non-coal mine safety fan which comprises a fan body, a mounting cylinder is arranged on the outer side of the fan body, clamping rods are fixedly connected to the two sides of the mounting cylinder, clamping sleeves are arranged on the outer sides of the clamping rods in a sleeving mode, locking sleeves are arranged on the outer sides of the clamping sleeves, a plurality of locking grooves are formed in the outer sides of the clamping sleeves, and releasing sleeves are arranged on the outer sides of the clamping sleeves. A locking frame is arranged on one side of the release sleeve, a locking rod is arranged in the locking frame, a sliding groove is formed in the clamping sleeve, a clamping block is arranged on the outer side of the clamping rod, a moving groove is formed in the outer side of the clamping rod, a movable groove is connected to one end of the moving groove, a positioning groove is connected to the other end of the movable groove, a release groove is formed in the inner side of the release sleeve, and a clamping groove is formed in the inner side of the clamping sleeve. According to the fan protection device, the safety protection of the fan is realized, the mounting stability of the protection plate is ensured while the protection plate is conveniently disassembled and assembled, and the stable use of equipment is ensured.

The utility model relates to a medical gauze wrapped cotton cushion dressing, and belongs to the technical field of medical dressings. The medical gauze wrapping cotton pad dressing comprises a cotton pad, the cotton pad comprises a cotton block and wrapping cotton gauze wrapping the cotton block, and the cotton pad further comprises a water-repellent film arranged between one surface of the cotton block and the wrapping cotton gauze; the medical gauze wrapped cotton pad dressing further comprises a gauze bandage used for fixing the cotton pad. According to the utility model, the water-repellent film is arranged between one surface of the cotton block and the cotton gauze, so that the surface of the dressing can be kept dry; and secondly, the dressing is compressed and plastically packaged, so that the problem of large occupied space during storage and transportation of the dressing is solved, and the vacuum plastic packaging also has a good antibacterial effect.

This invention belongs to the technical field of federated learning, and more specifically, relates to a federated image classification method and system based on visually guided attribute correction. The method includes: a server constructing a global model for image classification, comprising an image encoder, a text encoder, a cue learning module, and a gated text correction module, and distributing trainable parameters to each client participating in the federated learning; each client updating the parameters in its initial local model based on the trainable parameters distributed by the server, and training the updated local model using its local image dataset; and uploading the parameters from the trained local model to the server for aggregation to obtain the target global model. This invention addresses the shortcomings of existing attribute-assisted federated learning methods, such as static and rigid text attributes, inability to adapt to heterogeneous data with non-independent and identically distributed characteristics, and lack of cross-modal dynamic interaction mechanisms.

This invention relates to the field of power grid information security technology, specifically a method, system, device, and storage medium for detecting malicious software in the power grid Internet of Things (IoT) based on active learning. The method involves acquiring application samples from power grid IoT nodes, extracting static features, dynamic features, and power grid context information to form a multi-dimensional feature vector, and organizing these into data blocks according to timestamps. Classification uncertainty scores are calculated from an unlabeled sample pool, a detector committee is constructed to calculate consensus entropy, and the sample with the most information content is selected by combining the two scores and submitted for expert annotation. A random forest classifier is trained to build a detection model. The F1 score of the current data block is evaluated; annotation stops when a preset threshold is reached and is applied to the next data block. Model performance changes are monitored, and when performance degradation is detected, batch retraining, rolling back historical configurations, or incremental updates are performed based on the evolution of the threat environment. The method reduces annotation costs through sample selection and addresses the conceptual drift problem of the power grid threat environment through an adaptive update strategy.

This invention discloses a commutation method for a modular multilevel converter in quasi-two-level mode, comprising: determining the commutation stage and output current polarity, identifying the delayed control object, and determining the operation mode based on the phase voltage level switching direction; directly executing the operation of the non-delayed control object according to the determined operation mode and constructing two candidate paths for the delayed control object: direct operation and delayed operation; calculating the circulating current prediction value and the absolute error with the reference value corresponding to the direct operation and delayed operation respectively, and determining whether the delayed control object should operate directly or delayed based on minimizing the error; introducing a piecewise linear trajectory synthesis fine-tuning mechanism based on duty cycle, determining the delay time of the bridge arm submodule by calculating the duty cycle, and repeating the steps until the commutation process ends; this invention quickly completes the commutation of the upper and lower bridge arms by adding a delay, and at the same time, the piecewise linear trajectory synthesis mechanism based on duty cycle eliminates the steady-state error caused by the whole-cycle delay.

The present invention relates to a method for preparing a negative electrode active material, which includes mixing green coke particles, calcined coke particles, and a binder, and graphitizing the mixture through heat treatment to form an artificial graphite particle in the form of a secondary particle in which primary artificial graphite particles are bonded to each other, wherein an average particle diameter (D50) of the green coke particles is greater than an average particle diameter (D50) of the calcined coke particles.

The application provides a partition plate, a compressor pump body, a compressor and an air conditioner. The partition plate is arranged between two adjacent cylinders. The partition plate comprises an axial first end surface and an axial second end surface. The axial first end surface and the axial second end surface are oppositely arranged. A first heat insulation hole is arranged in the interior of the partition plate. The first heat insulation hole is located between the axial first end surface and the axial second end surface and does not contact the axial first end surface and the axial second end surface respectively. In the axial direction of the partition plate, the first heat insulation hole is opposite to a first area between a suction hole of the cylinder and an exhaust hole. According to the application, heat transfer from the high-temperature side of the cylinder to the suction side of the cylinder through the partition plate body can be inhibited, the suction temperature rise of the suction side of the cylinder is reduced, the refrigerant suction amount is increased, and the compressor performance is improved.

This invention relates to the field of radiation source identification technology, providing a method and apparatus for pulse stream segmentation and signal pattern recognition. The invention performs outlier detection on a radar pulse stream sequence to determine anchor points; segments the radar pulse stream sequence into multiple pulse subsequences based on the anchor points; extracts features from the multiple pulse subsequences to obtain original features; reweights the original features to obtain features to be classified; predicts the signal pattern of the features to be classified to calculate the classification loss, and iteratively updates the model to obtain a target recognition model, which is then used to perform signal pattern recognition on the segmented multiple pulse subsequences. This invention addresses the problem that outliers in radar pulse stream sequences cause performance degradation in radar signal pattern recognition algorithms.