64results about How to "Low computing resource requirements" patented technology

The invention discloses an ultrasonic positioning device for a four-rotor unmanned aerial vehicle. The device comprises the four-rotor unmanned aerial vehicle, an ultrasonic emission control module, an ultrasonic emission module, an ultrasonic receiving control module and a plurality of ultrasonic receiving modules. The ultrasonic receiving control module is connected with the ultrasonic receiving modules and fixed to the ground, one end the ultrasonic emission control module is connected to the output end of a flight control plate of the ultrasonic emission control module, and the other end of the ultrasonic emission control module is connected with the ultrasonic emission module. According to an ultrasonic positioning method for the four-rotor unmanned aerial vehicle, the ultrasonic receiving modules are adopted, every three ultrasonic receiving modules form one set for spatial positioning, the positioning result and positioning results of other sets are corrected mutually, positioning precision is improved, and errors are reduced. The ultrasonic positioning method for the four-rotor unmanned aerial vehicle has the advantages of being easy to operate, simple and accurate in calculation, low in computing resource requirement for a factual system and relatively low in cost, and can achieve the positioning precision of the four-rotor unmanned aerial vehicle more easily.

The invention discloses an offline commodity authentication method based on width learning and wide-angle microscopic images and belongs to the technical field of commodity authentication. The methoddisclosed in the invention comprises the following steps: wide-angle microscopic images of real and fake commodity samples are collected, static features of the wide-angle microscopic images are extracted to construct a tagged feature vector set, and the feature vector set is divided into a training set and a test set; the feature vector set is used for pre-training so as to build a width learningmodel; finally, the breadth learning model is transplanted to a mobile device; wide-angle microscopic images of commodities to be identified are collected, feature vectors of the wide-angle microscopic images are extracted, the feature vectors are input into a width learning model so as to perform true and false identification of the commodities; if the authentication of the commodities is incorrect, the feature vectors of misjudged true and false commodities are obtained; the acquired feature vectors are used to update a width learning model via an incremental learning method. The method canbe employed on the mobile device so as to realize offline detection, high accuracy can be achieved, and self-adaptive learning ability can be realized.

The invention discloses a width learning-based long-term target tracking algorithm. The algorithm comprises the following steps of: establishing a width learning system; and carrying out tracking on the basis of the width learning system and establishing an acceleration steady feature algorithm-based whole-image detection mechanism. According to the algorithm, width learning system-based long-termtarget tracking is carried out, width learning architecture is relatively shallow in layer and relatively low requirement is put forward for calculation resources, so that deployment can be carried out on low-end equipment without losing great precision. According to the algorithm, an obtained target tracking model is high in training speed, small in reconstruction cost and low in time cost, andhas great superiority for detection of deformation, rotation and shielding in target tracking process. According to the algorithm, the SURF algorithm-based whole-image detection mechanism is applied,so that target information can be rapidly obtained when targets re-appear under the condition that the width learning system judges that the targets are lost, and the target positions can be updated so as to ensure that the tracking effect is more stable, robust and reliable.

The invention discloses a monocular camera and millimeter wave radar external parameter calibration result accuracy evaluation method. A plurality of metal balls are arranged in a scene, after data information of the metal balls in all coordinate systems is converted, statistics and comparison calculation are completed, and four indexes describing the accuracy of an external parameter result are obtained, and two indexes are in a camera image plane and two indexes are in a radar point cloud space.

The invention discloses a safety helmet image recognition algorithm based on width learning. The safety helmet image recognition algorithm is characterized by comprising the following steps: S1, preparing data of a width learning system; S2, training a safety helmet image recognition model based on width learning; S3, a safety helmet image detection mechanism based on a width learning system: S31,reading the image test set obtained in the step S1 and the safety helmet image recognition model obtained in the step S2; S32, inputting images in the image test set into the safety helmet image recognition model to obtain an output value; And S33, marking a picture of a person without a safety helmet, and giving an alarm. The method is a safety helmet image recognition technology based on a width learning system, the width learning architecture is shallow in hierarchy, and the requirement for computing resources is low, so that the wide learning architecture can be deployed on low-end equipment without losing too much precision.

The invention discloses a video monitoring method, a video monitoring system and a video monitoring device. The video monitoring method includes enabling the front-end monitoring device to perform pre-analysis operation on acquired monitoring video stream data according to preset pre-analysis functions; identifying pre-analysis results of the pre-analysis operation and judging whether alarm conditions corresponding to the pre-analysis functions are met or not; responding to the pre-analysis result that the alarm conditions are met, acquiring upload monitoring data from the monitoring video stream data according to upload setup information, and transmitting the upload monitoring data to a back-end analysis platform; enabling the back-end analysis platform to intelligently analyze and process the upload monitoring data. According to the technical scheme, the video monitoring method, the video monitoring system and the video monitoring device have the advantages that the volume of monitoring data transmitted from the front-end monitoring device to the back-end analysis platform can be reduced, accordingly bandwidth resources occupied by monitoring data transmission can be saved, the data volume of the monitoring data analyzed by the back-end analysis platform can be reduced, and the requirement on computation resources of the back-end analysis platform can be lowered.

The invention provides a multi-agent based ordered charging method of an electric automobile. The method is characterized in that a home agent and a transformer substation agent coordinately work to service a power grid agent to handle with a large amount of complex charging requests in a distribution manner; and the charging behavior of the electric automobile is scheduled through a real-time greedy algorithm. According to the method, the current real-time energy demand information and the charging time arrangement of the electric automobile, rather than the information of the electric automobile to be connected to a power grid in future and a complex prediction method and the optimizing technology, are needed, so that the expansion is high; and the method is applied to the high permeability condition of the electric automobile. The charging process of the electric automobile is a constant-power continuous charging process through which a battery of the electric automobile is prevented from being frequently switched among the charging modes, and the influence of the charging schedule on the service life of the battery of the electric automobile can be reduced to the maximum. With the adoption of the method, the real-time charging optimizing effect of the electric automobile is effectively improved; the peak shaving and load shifting can be performed well while the user demand is met and the user charging cost is decreased, and the power grid stability is ensured.

The invention relates to a mobile robot autonomous navigation software framework and a navigation method. The software framework is divided into a user interface layer, an autonomous navigation layer,a system monitoring layer, an equipment layer and a bottom application layer. Each functional module operates according to an independent thread, and data interaction is carried out between the modules in a memory sharing mode. The software framework separates a sensing module, a planning module and an execution module, and can adapt to platforms in different driving forms, different sensor configurations and different tasks. Common abstract interfaces such as maps, paths, classifiers, tasks and path planners in a navigation system are defined in a standardized mode, and a common coordinate system and data structures related to sensors and functional modules are defined. The bottom application layer realizes thread, mathematical operation, log and hardware interface access; and the equipment layer realizes definition and basic operation of the functional modules, realizes a data acquisition function of common equipment, has relatively complete underlying functions, and can support agile development.

The invention discloses a rhythm-controllable Chinese and English mixed speech synthesis method and system, and belongs to the field of speech synthesis. The method comprises the following steps of: 1) extracting phoneme pronunciation duration, energy, pitch and Mel spectrum from Chinese and English texts and audios with rhythm marks as a training set, and learning a text coding representation aligned with the Mel spectrum in length; and 2) generating predicted energy and pitch through an energy and pitch prediction model to realize energy and pitch control. And 3) combining the predicted energy, pitch and text coding representation, outputting a synthesized Mel spectrum through a decoder, and synthesizing speech through a vocoder. According to the method, the rhythm pause information in the synthetic speech is better controlled by using a jump neural network encoder, the rhythm pronunciation information of each frame in the synthetic speech is finely controlled by using the predicted duration, energy and pitch, and the speech with richer rhythm change is generated. The method is completely completed by one model, and language distinguishing does not need to be carried out on the texts firstly.

The invention discloses a phishing website detection method based on web automatic testing and width learning, and belongs to the technical field of computer network safety. According to the invention, firstly, conventional feature extraction is carried out on the basis of url (Uniform Resource Locator) and html pages, then interactive feature extraction is carried out by utilizing a web automatictesting technology, and finally, width learning training is carried out by utilizing a preprocessed training sample after feature extraction, so that a phishing website is accurately and rapidly identified and detected, and network information and property safety of the public is protected.

The invention discloses a wind turbine generator single wake flow distribution simulation method based on CFD pre-calculation, and belongs to the technical field of detection. Modeling is carried outon unit wake flow fields at several different incoming flow wind speeds in advance, then CFD numerical simulation is carried out, and a unit wake flow field information database at discrete incoming flow wind speeds is obtained; wake flow field speed distribution of the unit at any incoming flow wind speed is obtained through an interpolation method based on the database. Different CFD simulationmethods can be selected according to different requirements for wake flow simulation precision in engineering, the method has high efficiency and high precision, and timeliness can meet actual engineering requirements.

An encoder (109) comprises a receiver (201) which receives a time domain audio signal. A filter bank (203) generates a first subband signal from the time domain audio signal where the first subband signal corresponds to a non-critically sampled complex subband domain representation of the time domain signal. A conversion processor (205) generates a second subband signal from the first subband signal by subband processing. The second subband signal corresponds to a critically sampled complex subband domain representation of the time domain audio signals. An encode processor (207) then generates a waveform encoded data stream by encoding data values of the second subband signal. The conversion processor (205) generates the second subband signal by direct subband conversion without converting back to the time domain. The invention allows an oversampled subband signal typically generated in parametric encoding to be waveform encoded with reduced complexity. A decoder performs the inverse operation.

The embodiment of the invention provides a timing control method and system, and relates to the technical field of the timing. The method comprises the following steps: a server receives a countdown request associated with intelligent equipment and sent by a client, when the interval between the time node of the countdown request and the current time is less than a preset first threshold, the server receives the current residual countdown amount sent by the intelligent equipment; when the current residual countdown amount is greater than zero, the server sends the countdown request to the intelligent equipment, thereby enabling the intelligent equipment to execute the countdown task. Through the timing control method and system provided by the invention, the requirement on the server computing resource is greatly reduced, and the server can support more amount of countdown task.

The invention discloses a pedestrian image retrieval method based on text description, which comprises the following steps: S1, constructing a pedestrian image retrieval model, and training the pedestrian image retrieval model to obtain a trained pedestrian image retrieval model; and S2, calculating a cosine similarity value between the image features and the text features by adopting the trainedpedestrian image retrieval model, and obtaining a pedestrian image retrieved based on text description according to the cosine similarity value. According to the invention, the problem that the intra-class difference of pedestrian features in a pedestrian image retrieval task based on text description is too large and too small in the prior art is solved.

The invention provides an immunohistochemical cell image cell nucleus segmentation and counting method, which comprises the following steps: acquiring a pathological cell image through a preset input data source; determining a display image on an image display panel according to the pathological cell image and an input data source; determining a cell nucleus segmentation area on the display image through graph drawing; performing color correction and noise reduction on the cell nucleus segmentation region, and determining an initial cell image; processing the initial cell image, obtaining a negative area mask, a positive area mask and a cell nucleus contour, classifying cell nucleuses detected in the initial cell image according to the cell nucleus contour, the negative area mask and the positive area mask, and determining classified cell nucleuses; and counting the classified cell nucleuses, determining a diagnosis index according to the classified cell nucleuses, and optimizing a display result according to the diagnosis index.

The invention discloses a method for quickly acquiring comparison result data of a target genome region. The method comprises the following steps: acquiring a reference genome sequence file and coordinate information files of all genes by using a public genome database on the basis of sample original sequencing data, and constructing a reference genome index file and a chromosome index file; constructing a mapping relationship between the sequence line number of the sample original sequencing data and genome comparison coordinates, and quickly reconstructing sample original sequencing data of a target gene sequence by using the mapping relationship; and carrying out sequence comparison by utilizing the chromosome index file and the sample original sequencing data of the target gene sequence to obtain an original comparison data file of the target gene sequence, and carrying out sequencing and duplicate removal to obtain final comparison result data of a target genome region. The method has the characteristics of simplicity in deployment, convenience in operation, high efficiency, high throughput and wide application range. Compared with an original secondary data BAM file, the obtained result basically has no information loss.

The invention provides a virtual laser radar system and method based on mutual early warning of fans, and the system comprises: a wind field deployment data module which obtains the front-back relation of fans which are arranged at intervals in a main wind direction in a wind field; a wind speed transmission characteristic module which is used for acquiring the transmission characteristic of the regional wind speed of the wind field in the main wind direction; a wind field operation data module which is used for acquiring actual operation data of a wind field; and a neural network module whichcaptures the wind speed relationship between the fans according to the front-back relationship of the fans arranged at intervals in the main wind direction in the wind field, the transmission characteristics of the wind speed in the main wind direction and the wind field operation data so as to replace the physical laser radar to predict gust.

The invention discloses an octree map construction method and system. The method comprises the following steps of determining a key frame and a mobile robot pose under the key frame by adopting a laser map optimization SLAM algorithm according to laser radar and odometer data, collecting environment point cloud information, and converting the point cloud data under a camera coordinate system into a world coordinate system, splicing the point cloud information after coordinate conversion to obtain a three-dimensional point cloud map, based on the three-dimensional point cloud map, conducting loopback detection and back-end optimization on a laser map optimization SLAM algorithm, and acquiring an updated mobile robot pose and an updated three-dimensional point cloud map, filtering the updated three-dimensional point cloud map, and removing outliers of the updated three-dimensional point cloud map, and converting the filtered three-dimensional point cloud map into an octree map. Compared with a point cloud map, the octree map constructed by the method is easy to store, smaller in occupied memory space, flexible and convenient, and can be directly used for navigation.

The invention provides an aircraft final assembly production line productivity prediction method based on a deep neural network. A resource configuration scheme and historical capacity information areuniformly expressed in a data sequence form with time sequence information and are used as neural network input; a deep belief network (DBN) and a back propagation (BP) algorithm are used, and training of the aircraft final assembly production line productivity prediction model is completed by combining an unsupervised training mode and a supervised training mode, so that change prediction of theaircraft final assembly production line productivity in a period of time in the future is realized.

The invention provides a heart sound diagnosis system with deep learning and a low-difference forest mixed. The system comprises a preprocessing module configured to preprocess acquired heart sound signals, and normalize, filter and down-sample the heart sound signals in sequence; a data conversion module configured to extract audio features of the down-sampled data to generate second-order spectrum data; a deep learning module configured to perform feature extraction on the second-order spectrum data by using a trained deep learning model; and a feature classification module configured to classify the extracted features by using a trained low-difference forest classifier to obtain a classified diagnosis result. The system has low requirements on professional equipment and professionality of personnel, and is portable and universal.

The invention discloses a visual obstacle detection method based on an FIRA platform, and the method comprises the steps: image preprocessing: carrying out the graying processing and corrosion expansion operation of an input image; extraction of a longitudinal obstacle: filtering out pixels to which the obstacle belongs from the preprocessed image, and extracting the position of the obstacle corresponding to each column of pixels by counting the number of each column of pixels; obstacle segmentation: performing column direction scanning operation according to the extracted entity coordinates corresponding to the pixels in each column, judging whether the columns belong to the same obstacle or not, and judging the shielding relationship between adjacent obstacles; exception processing: performing exception processing on the result of preliminary obstacle segmentation, extracting side-by-side obstacles with overlong width, then segmenting the midpoint into two obstacles, extracting obstacles at the edge of the picture, and determining the positions of the obstacles through side edge vertexes; and outputting a result. According to the method, environmental geometric information and a camera imaging principle are fully utilized, and the computing resource requirement is greatly reduced under the condition that the precision requirement is met.

The invention provides a self-assembly-colloidal-crystal computer simulation method based on heat-treatment assistance and a preparing method of colloidal single crystals. According to the self-assembly-colloidal-crystal computer simulation method based on the heat-treatment assistance, a physical model for optimizing colloidal crystal technological parameters with the Monte Carlo algorithm is built, and technological parameters such as the optimal self-assembly temperature are simulated and predicted through the physical model; meanwhile, the glue evening method is combined, a plurality of other self-assembly parameters of micro-nano ball crystals are controlled and assembled, and large-area colloidal single crystals are obtained. A theoretical model under the condition that liquid evaporation of the colloidal single crystals is considered is built; based on the theoretical model, computer simulation is introduced, and the efficiency of the self-assembly colloidal crystals is greatly improved; the algorithm of the model is efficient and convenient and fast, quite-large computing resources are not required, the obtained result and the experimental result are highly identical, and the quite-good experiment instruction value and market application prospects are achieved. According to the computer simulation method, multiple codes are specifically adopted in the method; the method for preparing the colloidal crystals is suitable for assembling micro-nano balls with the large-range size, and can also be suitable for assembling different kinds of micro-nano balls.

The invention relates to a method and a device for detecting a human reference genome virus integration site, and belongs to the technical field of gene detection bioinformatics analysis. The method comprises the steps of genome alignment, sliding cutting, short sequence alignment and integration. According to the method, reads which are not compared with a human reference genome and a virus genome at the same time are subjected to sliding cutting, the cut sub-sequences (or short sequences) are compared again, through clustering, correlation and covariance processing of the comparison positions and the sequence of a certain reads segmentation sequence, all highly possible comparison positions can be listed in the analysis process, reads for simultaneously comparing human reference genomesand virus genomes can be accurately found out and accurately positioned, and the error range is within 3bp. The method is low in calculation resource requirement and high in operation speed, and has relatively high practical value.

The invention provides a monitoring method for a ship passing through a special area, and the method can carry out the cleaning, conversion and correlation storage of massive ships and ship AIS data, and draws a more precise ship track; the method also includes analyzing AIS data, and providing more accurate ship state judgment; carrying out electronic processing and division on the global special area; providing risk early warning for the ship to pass through a special area in real time, and solving the pain point of post declaration at present. According to the invention, global ship mass AIS data is cleaned and sorted, electronic processing is carried out on a shipping global special area, and timely and accurate ship special area passing monitoring is provided. In addition, a special area is reduced through circumscribed rectangle processing, and the requirement for computing resources of traversal data judgment is lowered.

The invention discloses a confrontation-based bidirectional consistency constraint medical image registration algorithm. The algorithm comprises the following steps: acquiring medical images of CT and MRI modals, and carrying out standardized preprocessing on the medical images; constructing a deep learning neural network, including construction of a generator and construction of a discriminator; randomly selecting a training sample set, and performing adversarial training on the deep learning neural network constructed in the step S2; and inputting the to-be-registered medical image pair into the trained deep learning neural network, and outputting the registered image pair, so that the problems of few medical image registration samples, difficulty in multi-modal registration and difficulty in similarity measurement selection at present are solved, and a high-precision registration effect is achieved. According to the medical image registration method, the medical image registration method and the medical image registration system, for example, registration of a preoperative CT image and an MRI image of the same patient, algorithm training can be completed without providing a large number of sample sets and sample labels, meanwhile, operation stability is guaranteed, and a high-precision medical image registration image pair can be automatically obtained.

The invention relates to the technical field of production plan scheduling, in particular to a collaborative cloud production scheduling method and system based on deep reinforcement learning. The method comprises the following steps: a client inputs scheduling demand information to a scheduling decision module; edge equipment acquires production state data of each production line in an intelligent workshop in real time, and uploads the production state data to an edge cloud for primary processing of the data; the edge cloud uploads the primarily processed data to a core cloud for data fusion; the scheduling decision module receives the fusion data and the scheduling demand information, and generates a scheduling strategy corresponding to each production line by using a deep reinforcement learning algorithm; the scheduling decision module transmits scheduling strategy information back to the edge cloud, and the edge cloud realizes production scheduling control on machines of each production line; and in a production process, an abnormal event monitoring module monitors abnormal data in the production process of the intelligent workshop in real time and gives an alarm feedback. According to the invention, the problem that current workshop production scheduling efficiency is low can be solved.

The invention provides a multi-agent based ordered charging method of an electric automobile. The method is characterized in that a home agent and a transformer substation agent coordinately work to service a power grid agent to handle with a large amount of complex charging requests in a distribution manner; and the charging behavior of the electric automobile is scheduled through a real-time greedy algorithm. According to the method, the current real-time energy demand information and the charging time arrangement of the electric automobile, rather than the information of the electric automobile to be connected to a power grid in future and a complex prediction method and the optimizing technology, are needed, so that the expansion is high; and the method is applied to the high permeability condition of the electric automobile. The charging process of the electric automobile is a constant-power continuous charging process through which a battery of the electric automobile is prevented from being frequently switched among the charging modes, and the influence of the charging schedule on the service life of the battery of the electric automobile can be reduced to the maximum. With the adoption of the method, the real-time charging optimizing effect of the electric automobile is effectively improved; the peak shaving and load shifting can be performed well while the user demand is met and the user charging cost is decreased, and the power grid stability is ensured.

The invention provides a network adaptive semi-precision quantification image processing method and system. The method comprises the following steps: collecting image data of edge computing equipment;inputting the image data into a pre-established depth residual convolution quantization network for image processing for processing to obtain the category of a target corresponding to the image data,positioning and the category of pixels in the image; wherein the deep residual convolutional quantization network is obtained by training based on a deep network adaptive semi-precision quantizationmethod, and the semi-precision quantization is carried out by adopting a floating-point number with a half digit. Semi-precision adaptive quantization is carried out on the deep residual convolution quantization network, the size of the deep residual convolution quantization network can be reduced on the premise that the precision is guaranteed, and the requirement for computing resources is lowered.

The invention discloses a thyroid nodule detection method based on Transform, and relates to the technical field of image processing, and the method comprises the steps: obtaining a to-be-detected ultrasonic image of a thyroid region, carrying out the image preprocessing of the obtained to-be-detected ultrasonic image, inputting the to-be-detected ultrasonic image into a nodule detection model which is obtained through the training based on a Transform network in advance, and obtaining a nodule detection model; according to the output of the nodule detection model, determining the position and type of the nodule in the to-be-detected ultrasonic image, and completing the detection of the nodule in the to-be-detected ultrasonic image, the nodule type being used for indicating that the nodule is a benign nodule or a malignant nodule; the method can automatically complete positioning and classification of nodules, is high in automation degree and good in objectivity, does not need to construct a dense Anchor Box, does not need to use NMS complex post-processing operation, is easy to implement, and has low requirements for computing resources.