67results about How to "Low computing power" patented technology

The invention discloses a double-spectrum night vision instrument vehicle-mounted system and a double-spectrum fusion design method. The system comprises an infrared double-spectrum imaging die set used for imaging of a visible spectrum and an infrared spectrum and for collection, processing, fusion and transmission of the image data, an image receiving system used for completing the receiving of the image data, a display terminal used for displaying the images for an observer, and a control terminal used for controlling the function and on-off of an imaging mode. Due to the adoption of the system and the method, the problem that coaxial imaging of the double spectrum image is slight bad is solved, the machining precision requirement of the imaging system is reduced, more clear images are provided for forward observation for a train; a fusion method based on a scene moving object provides a reference source for the double-spectrum imaging system, the computation load is reduced, the registering reliability is improved, the imaging on the hardware is convenient, and the real-time performance is possessed; a weight fusion method based on multi-scale bilateral filtering is simple and feasible, improves the imaging detail information and reduces the calculation ability.

A method for capturing a sequence of video images, by means of an imager including an estimation of the parameters of a model of global motion between successive images. In one embodiment, the method comprises a measurement of local motions on edges of the images, and the estimation of the parameters of the global motion model is performed using the result of the measurement of local motions on the edges of the images.

The invention provides an Internet information putting channel optimizing system based on distributed computing. A data collecting module collects user behaviors; a data preprocessing module carries out data cleaning, integration and reduction, and simplifies and standardizes the collected user behavior information; a training module carries out iterative operation on the data in a training set by a class E-M algorithm, and parameters in a probability cumulative model are obtained; then, data in a test set is substituted into the probability cumulative model, so that the prediction on the contribution of each putting channel and the prediction on whether users are converted or not are completed; and websites or recommending types with a high contribution degree are subjected to information putting, and the information is put to users most possible to generate conversion. Meanwhile, an Hadoop platform is used for distributed computing, and complicated computing consuming a great number of resources is distributed onto multiple nodes for computing, so that the multi-node parallel processing is realized.

A method for compressing image includes selecting at least two statistic points from image, judging whether color number sum of two selected statistic points is greater than preset color number threshold or not and applying destructive compression algorithm to carry out compression on image if it is or otherwise applying nondestructive compression algorithm to carry out compression on image. The compression device used for realizing said method is also disclosed.

The invention discloses an arhythmia testing method for aiming at electrocardiogram data by means of a binary neural network. The method comprises the steps of firstly acquiring training model data, and preprocessing the data before training; constructing a set of full-precision convolutional network model, and inputting data for training, adjusting a network parameter for obtaining a relatively high effect; referring to the obtained full-precision models for constructing a binary network model, inputting the data for training, performing fine adjustment on a model parameter, and utilizing a Stop-BN training method in training for improving a training effect; using the trained full-precision model as a teacher model and using the untrained binary model as a student model, and performing distillation training on the student model by means of the teacher model, thereby obtaining a training effect better than that in directly training the binary network. The method performs discriminationon atrial fibrillation and can effectively reduce operation memory and operation time. The trained network model realizes relatively high effect for reducing precision loss caused by binarization.

The invention relates to a low-cost low-power sound source orientation method. A hardware system of the sound source orientation method comprises a single-chip microcomputer, multiple microphones, multiple operational amplifiers and one communication interface chip, wherein the microphones are connected with the AD interface of the single-chip microcomputer respectively through the operational amplifiers, and the single-chip microcomputer is connected with a computer through the communication interface chip. The sound source orientation method comprises the following specific steps: step one, consistency screening is performed on the multiple microphones; step two, four microphones subjected to the screening in the first step are selected and are arranged in the form of a cross array, the microphones on diagonal lines of the cross array are paired, and a sound source is oriented by use of a sound source orientation method of an arrival time difference; and step three, a calculation result is output to a communication port. The sound source orientation calculation method brought forward by the invention is simple and highly efficient, has quite low requirements for an operation capability and resources of a processor and can meet the requirement for realizing sound source orientation at a low cost.

The invention relates to a method and a device for handwriting track detection, a method and a device for handwriting track standardization, a method and a device for handwriting track online-identification, and a method and a device for abnormal radical collection. According to one embodiment of the invention, the method for detecting the character handwriting track is provided. The method comprises the following steps: first judgment is made on each stroke of a handwriting track according to the geometric features of the stroke based on predetermined abnormal radicals to judge whether the stroke is an abnormal radical; and if the stroke is not an abnormal radical, second judgment is made to judge whether the stroke is part of an abnormal radical and whether the combination of the stroke and an adjacent stroke is an abnormal radical, wherein the second judgment is performed on the basis of the geometric features of the stroke and the geometric features of the adjacent stroke.

The invention discloses a scanning method for an infrared touch screen. The scanning method includes: starting infrared light emitting and receiving diodes of the touch screen to perform scanning at a preset scanning angle during responding to at least one touch control operation of a touch article on the touch screen so as to acquire information of shielded light beams between the infrared light emitting and receiving diodes of the touch screen, wherein the infrared light emitting and receiving diodes includes infrared light emitting diodes and infrared light receiving diodes; calculating candidate touch point zones based on the information of the shielded light beams, wherein the number of angles of the shielded light beams of the candidate touch point zones is greater than a first threshold, and the first threshold is less than a preset number of scanning angles; determining each candidate touch point zone having the number of privately-owned light beams greater than or equal to a second threshold as a true touch point zone, wherein the number of the privately-owned light beams of the each candidate touch point zone is the number of shielded light beams passing the candidate touch point zone.

The invention discloses a fault diagnosis method for a mine fan. The fault diagnosis method comprises the following steps: utilizing SCADA to detect a real-time state of fan operation; adopting a waycombining machine learning and deep learning to train data monitored under various working states of the fan; and adopting a GBDT-CNN model to perform fault diagnosis on fan equipment. After determining and judging the fault position of the integral equipment, the diagnosis method determines specific fault degree including light fault degree, medium fault degree and heavy fault degree, of the device. A worker is reminded to select a proper equipment maintenance way while cost is saved as much as possible. Fault diagnosis of equipment is performed in step, and specific fault judgment is not performed while faults of the equipment are judged at the first step, so that diagnosis efficiency is improved while calculation ability of a computer is calculated. Relatively easy-to-capture relation exists between attributes of monitored data and judged fault types. A GBDT integrated learning algorithm is adopted, so that an accuracy rate is relatively high.

The invention relates to a single-trial electroencephalogram P300 component detection method based on the folding HDCA algorithm. A testee is made to watch a series of RSVP image sequences, the electroencephalogram signals of the testee are acquired, the electroencephalogram signals corresponding to all images are divided into a plurality of time windows, multi-lead signals in one time window and f time windows before the time winder form a group of new lead signals, space weight computing and time weigh computing are conducted to obtain an interest score, the interest score is compared with a set threshold value, a target image is judged, and a target image result is output. By correlating the electroencephalogram signals of the testee at the current moment with electroencephalogram signals at previous moments, the dimension of multi-lead electroencephalogram signals is reduced to one, the influences caused by the change of the incubation period and peak value of the P300 component along with the physiological status of the testee, target probability and target meaning in actual use are effectively reduced, the electroencephalogram P300 component is effectively extracted, and then the target image is determined.

The present invention relates to an intelligent terminal and discloses an intelligent terminal data interaction system and method. In the present invention, the system comprises a server-side apparatus and intelligent hardware of at least one terminal side, and the intelligent hardware comprises a wearable device or Internet-of-things terminal in which an intelligent operating system can not be installed; each piece of intelligent hardware is separately configured with a unique routing address; application pages are stored in addresses of multiple applications in the server-side apparatus; the intelligent hardware sends an address of a required application to the server-side apparatus; the server-side apparatus sends an application page stored in the address to the intelligent hardware of a corresponding routing address; and the intelligent hardware displays and stores the returned application page. In the present invention, operations are migrated onto the server-side apparatus, so as to reduce local operations and storage demands; and the server-side apparatus manages intelligent hardware and applications of different terminal sides by using the address and / or routing, and data between the intelligent hardware of the different terminal sides and different applications can perform free cross-platform interaction on a server.

The present invention belongs to the technical field of the internet of things, and discloses a system and a method for device safety access of a digital home network. A device authenticator is configured to perform safety authentication of all the devices. The device authenticator is added when the digital home network is constructed to perform safety authentication of all the devices. The deviceauthenticator comprises a control point and a sensor device, the control point is taken as a device for safety access to prevent from malicious access of an external device; a two-factor authentication method, the authentication of a control node is changed to authentication of an authentication server, and the device authentication server taken as a new UPuP device is introduced and called UPServer so as to change the authentication and registration flow of the device.

The invention discloses a network bypass device and a processing method thereof. The network bypass device comprises a controller unit, a driver unit and an actuator unit, wherein the controller unit comprises a main chip and multiple auxiliary chips, the main chip is connected with the processing end and exchanges data with the processing end, and the auxiliary chips exchange data with the main chip; the driver unit is connected between the controller unit and the actuator unit and used for being connected to the auxiliary chips through multiple network bypass channels and driving the actuator unit to work according to data issued by the controller unit; the actuator unit is connected with the network end through a difference line and is used for processing network difference signals interacting with the network end. The network bypass device has the advantages that by means of work division operation of cascading bypass controllers, the number of the network bypass channels is increased, and upper software and parameter configuration can be uniformly controlled. One controller takes charge of management of all data information, the data can be obtained and issued in a centralized mode, and operation control and fault location are easy.

The embodiment of the invention provides a data transmission, network distribution and management method, device and system and a storage medium. In the embodiment of the invention, the Bluetooth device or module working in the broadcast mode encrypts the data by using the preset key and then broadcasts the encrypted data outwards, and correspondingly, the server device decrypts the encrypted data by using the locally pre-stored key. In the whole process, the data is encrypted and transmitted, and the key is not transmitted in the air, so that the security of the data is ensured. Besides, when the Bluetooth device or module works in the broadcast mode, the broadcast mode has low requirements on the operational capability and hardware resources of the Bluetooth device or module, so that the implementation cost of the Bluetooth device or module can be reduced.

The invention discloses a method and device for measuring the proportion of impurities contained in harvested grains of a combine harvester, and the method comprises the steps: enabling an industrialcamera, illumination equipment and damping equipment to be installed at a corresponding position, needing to measure the proportion of impurities contained in harvested grains, of the combine harvester, and adjusting the shooting interval and exposure time of the industrial camera; distinguishing grains in the acquired images from other impurities by utilizing a CPU (Central Processing Unit) through an image processing algorithm, and calculating the proportion of impurities contained in the harvested grains; and providing a display device, receiving specific gravity information of impurities contained in the grains through a parallel interface, and displaying the specific gravity information in real time. According to the method and device for measuring the specific gravity of the impurities contained in the grain harvested by the combine harvester, the grain and the impurities can be effectively distinguished, specific gravity information of the impurities contained in the grain is obtained, and the purpose of real-time and accurate monitoring can be achieved.

The invention relates to the field of wireless communication, and discloses a combined smart terminal, which can lower the cost of wireless mode updating or transmission of a network of different modes. The smart terminal comprises a sub-phone and a host phone capable of working independently, wherein the sub-phone in the independent state has the most basic voice communication and short message sending / receiving functions; and the host phone can be used independently as a multimedia terminal. The host phone has an internal cavity matched with the sub-phone in size, and the sub-phone can be inserted into the cavity completely. An opening extending to the cavity is processed on the casing of the host phone, and the position of the opening corresponds to that of the speaker of the sub-phone inserted into the cavity, so that the opening can be used for conducting the sound from the speaker of the sub-phone. After being inserted into the cavity of the host phone completely, the sub-phone is connected with the host phone through an electric interface to enable the sub-phone to achieve voice and data communication function as a wireless modem under the control of the host phone.

The invention discloses an Internet of Vehicles system, an intelligent driving control method and equipment thereof. The system comprises a sensor layer, a control layer, a coordination layer and a management layer, the sensor layer is used for packaging and transmitting sensor data acquired by a vehicle end to the management layer; the control layer is used for carrying out basic data transmission with the management layer, and the basic data is obtained from the sensor data; the coordination layer is used for performing fusion data transmission with the management layer, and the fusion data is determined by the basic data according to the vehicle intelligent driving service type; the management layer is used for service data transmission between a vehicle end and a cloud end, and the service data comprises at least one of the sensor data, the basic data and the fusion data. Through the layered setting mode, each level can be independently upgraded according to the intelligent driving service type as required, and a more convenient condition is provided for subsequent function upgrading of the vehicle.

The invention provides a sensor quantization method for a space optics effective loading force disturbance simulation source actuator and belongs to the technical field of testing space platform micro-vibration environmental adaptability through space optics effective loads. The sensor quantization method aims to realize the space optics effective loading force disturbance simulation source actuator. The method comprises the step one of determining a control loop type, the step two of determining a sensor type and the sensitivity, the step three of determining the control range and the minimum control step size, the step four of determining the sensor quantization bit width, and the step five of carrying out gain compensation on sensors of a loop. The method achieves the effect of sensor quantization of the space optics effective loading force disturbance simulation source actuator, normalization processing and resolution setting of multi-sensor quantization of the loop are controlled, and the situation that due to the fact that in a digital controller, a large number of division operations are guided into the sensor normalization processing, the operation performance and the control precision of the controller are reduced.

The invention relates to a method for adjusting a rotor position of an electrically commutated motor, in particular a brushless DC motor, in which the phases of the motor are actuated after determination of the rotor position via a block commutation. The invention is characterized in that the actuation of the motor phases is switched from block commutation to time-controlled commutation in order to adjust the rotor position with high precision.

The invention relates to an LED array animation rendering display control method and device. The method comprises the following steps that a pre-established LED animation model is obtained, wherein the LED animation model comprises a plurality of renderer objects, and each renderer object represents a continuous one-dimensional color subspace colored by a plurality of coloring points; according tothe animation actual frame rate requirement, the renderer object is scheduled in the LED animation model for rendering at the rendering moment of each frame, rendering of the one-dimensional color subspaces is completed according to the feature data and coloring configuration of the renderer object, color superposition based on the coloring points is conducted on all the rendered one-dimensionalcolor subspaces, and the color rendering space is obtained; and the color rendering space is mapped to an LED array. According to the technical scheme, various flexible animation display effects canbe obtained, and hardware cost is low.

The invention relates to an infrared imaging non-uniformity intelligent calibration and correction method and system, and belongs to the technical field of infrared imaging systems. The method comprises the following steps: firstly, carrying out two-point correction on a currently collected infrared original image; inputting the corrected image into a trained deep convolutional neural network for non-uniformity noise detection; and if the detection result is that the non-uniform noise is contained, starting a baffle switch of the infrared detector, and resetting correction parameters of the infrared detector. Compared with the prior art, the method has the advantages that the deep convolutional neural network is used for intelligently detecting the non-uniformity noise, and the problem that a radiation blocking piece in an infrared imaging system needs to be periodically and passively operated and parameters of an infrared detector need to be reset in an existing calibration-based non-uniformity correction method is solved; therefore, the operation frequency of the radiation separation blade of the existing infrared imaging system is greatly reduced, and the probability of accidents caused by interruption of image acquisition and untimely detection is reduced.

An APL (103), by way of a detection unit (202), detects that upon another core that is different from a specific core that executes specific processing, specific software among a plurality of software has functionality equivalent to specific processing, and that other software that is different from the specific software is being executed. After detection, the APL (103) extracts requirement values of computing capabilities of the specific software and the other software from a software capability requirements table (201) by way of an extraction unit (203). Using the extracted requirement values, the APL (103), upon evaluation by a surplus evaluation unit (204) that the total of the requirement values of the computing capabilities of the specific software and the other software is less than or equal to the computing capability value of the other core, assigns the specific software thereto by way of an assignment unit (205). After assignment, a PMU (102) causes a DSP (104) to be paused by a pausing unit (206).

The invention discloses an embedded instrument code disc character automatic identification system and method. The automatic identification system comprises an embedded microprocessor, an image sensor, a camera module, a TTL-USB to serial port module, a storage module and a power supply module. According to the invention, instrument code disc images under different angles and illumination conditions are adopted through the camera module, and image character areas are positioned and segmented; the obtained single code disc character size is normalized, and image shift processing is performed by using a data enhancement algorithm to obtain a whole character sample set and a double-half character sample set; a lightweight convolutional neural network is constructed, and the whole character sample set and the double-half character sample set are input into the neural network for training in a classified manner; and the trained network model is quantized, the quantized network model is deployed to a machine vision system, a single code disc character region is positioned, classification identification is carried out on single code disc characters in the region, and an identification result is displayed. Water level data monitoring and an image identification technology are combined, automatic identification of instrument characters is achieved, and the identification precision is high.

The invention relates to a marine radar rain and snow clutter suppression algorithm which comprises the steps of reading scanning line data, reading rain and snow clutter suppression parameters, establishing a sliding window matrix I and a sliding window matrix II, respectively calculating the average value of the amplitude in the sliding window matrix I, and performing weighted calculation according to the average amplitude value in the sliding window matrix I and the average amplitude value in the sliding window matrix II to obtain rain and snow clutter weighted amplitude values of all points of the current scanning line, and performing calculation to obtain the amplitude value of the current scanning line after rain and snow suppression. Through the algorithm, the rain and snow clutter estimation value is closer to the echo characteristics of the actual rain and snow clutter, the amplitude of the target echo is reserved, the rain and snow clutter can be inhibited as much as possible, and the influence on the target can be reduced to the minimum.