105results about How to "Real-time recognition" patented technology

The invention provides a plate structure impact load positioning method based on a distributed fiber bragg grating sensing network, and belongs to the field of impact monitoring of structure firmness monitoring. The method includes the steps of firstly, arranging the distributed fiber bragg grating sensing network; secondly, collecting an impact response signal; thirdly, extracting feature information and analyzing sensitive characteristics of an FBG sensor; fourthly, positioning the position of an impact point through the association algorithm of the maximum peak amplitude ratio of the impact response signal and position information. The positioning algorithm is simple, rapid, convenient and reliable and is high in instantaneity.

The invention relates to the field of computer visions, and proposes a face recognition method based on deep learning in a monitoring environment. The method comprises the following steps: building aface feature database, and performing feature extraction and feature comparison on faces in a video, thus face detection and face recognition are achieved based on an MTCNN network and a Sphereface network respectively; and by adding affine transformation, increasing the diversity of a data set, network clipping, network sparsification and layer-by-layer quantization, the problems of low recognition accuracy of a traditional face recognition method, differences in recognition accuracy between East and West faces, and excessive bandwidth required for the face recognition technology based on deep learning are mainly solved. Experiments show that the face recognition method proposed by the invention can be used in the monitoring environment and has higher recognition accuracy.

The invention belongs to the technical field of agriculture, in particular relating to a DSP (Digital Signal Processor)-based rice weed recognition system and method. The DSP-based rice weed recognition system consists of an image acquiring module, a rice weed recognition module and an image displaying module, wherein the image acquiring module and the image displaying module are connected with the rice weed recognition module respectively by virtue of an I2C (inter-integrated circuit) bus and a BNC (Bayonet Nut Connector) data bus; the image acquiring module consists of a CCD (charge-coupled device) camera supporting high-definition image and an SAA7115 decoding chip of Philips, the image displaying module consists of a Gadmei PL7006 liquid crystal television used for displaying a television signal, and the rice weed recognition module consists of a DM642DSP image processing chip, and expanding SDRAM (synchronous dynamic random access memory) and SAA7121 (system application architecture 7121) decoding chips. According to the invention, the real-time recognition of weeds in rice fields can be realized, and the system miniaturization and portability can lead the system to be capable of being flexibly arranged on rice planting field equipment, and improving the practical application range of the system.

The invention provides a food material freshness identification method and device based on deep learning, a refrigerator and a medium. The method comprises the following steps that food material imageinformation is collected; according to the food material image information and a pre-trained food material freshness identification model, food material freshness identification result information isdetermined; and on the basis of deep learning, the food material freshness identification model is trained and updated by taking the food material image information collected in the food material freshness identification process as a sample according to a preset period. According to the technical scheme, the identification accuracy is relatively high, real-time identification can be achieved, optimization and improvement are carried out in the application of the food material freshness identification model, so that the application range of food material freshness identification is improved, and the method and the device can be applied to complex scenes.

The invention discloses an outdoor tomato identification method based on subsection threshold image segmentation and light spot identification. The method includes: after normalized color difference Cd and color component ratio rGB are calculated, subsection threshold image segmentation is utilized to realize tomato non-light-spot area segmentation Rr; a light spot area segmentation algorithm is used to extract a candidate tomato light spot area Rb, and Rr and Rb form an initial tomato area Rt; area marking and denoising are performed to Rt; calculating area At of the initial tomato area and area Ab of the candidate tomato light spot area; in the same tomato area, if Ab is smaller than one third of At, Rb is reserved at the tomato area, otherwise Rb is reckoned as a background light spot area and removed, and only Rr is reserved; and area marking and denoising are performed to the finally obtained tomato area. Subsection threshold image segmentation and light spot identification are combined to overcome influences of natural light variation, and real-time fruit identification under different lighting conditions is achieved.

The invention discloses a method for recognizing a satellite communication signal modulation way. According to the scheme, the method comprises the following steps: 1, filtering a received signal to obtain a signal to be recognized; performing power spectrum smoothening, carrier frequency estimation and coherent demodulation on the signal to be recognized to obtain a demodulated signal; 3, estimating the code element rate of the demodulated signal; 4, determining whether the received signal is a binary phase shift keying signal or not by using a square spectrum; 5, determining whether the received signal is a quaternary phase shift keying signal or a hexadecimal quadrature amplitude modulation signal by using a signal quartic spectrum, carrier frequency and the code element rate; 6, determining whether the received signal is an octonary phase shift keying signal or not by using a signal eightfold spectrum; and 7, determining whether the received signal is a hexadecimal amplitude phase shift keying signal or not by using a signal duodenary spectrum, the carrier frequency and the code element rate. The method has high recognition accuracy, and can be applied to real-time monitoring of satellite communication signals. Through adoption of the method, totally-blind real-time recognition of the satellite communication signal modulation way can be realized.

The invention discloses a real-time identification method for a non-cooperative target in a shore-based radar sea clutter measurement area. The method comprises the following steps: (1) constructing adata receiving system based on an AIS(Automatic Identification System); (2) calculating a sea clutter measurement radar beam irradiation area; (3) establishing a non-cooperative target identificationmethod; and (4) determining whether a non-cooperative target is in the radar beam irradiation area. According to the method based on the AIS, the AIS longitude and latitude information provided by the non-cooperative target is used according to the irradiation range of radar wave beams, the non-cooperative target is identified in real time, and state information including time, latitude and longitude of the non-cooperative target can be displayed and extracted in real time, and further stored and managed.

The invention discloses an on-line identification method for steering characteristics of drivers. The on-line identification method comprises the following steps: S1, collecting steering data of cars when a plurality of drivers under test drive the cars and steer; S2, analyzing dependency among the steering data collected in the step 1 to obtain characteristic item data capable of describing the steering characteristics of the drivers; S3, clustering the characteristic item data by using a clustering algorithm, and enabling each characteristic item data class to be corresponding to a drive steering type; S4, establishing an off-line identification model by a BP (back propagation) neural network, extracting a weight value and a threshold value, and further establishing an on-line identification model; S5, collecting target steering data of the cars when the drivers under test drive the cars and steer, and identifying the steering characteristics of the drivers under test by the on-line identification model. According to the on-line identification method for the steering characteristics of the drivers, disclosed by the invention, in allusion to the problem that the operations of the drivers are not matched with car responses in the steering process of the cars, the on-line identification model is established, therefore the types of the steering characteristics of the drivers can be accurately identified in real time.

The invention discloses a fast face angle recognition method based on deep learning. Face angle recognition relates to rotation angles of a head around three axes in a three-dimensional vertical coordinate system, and a face angle change can cause partial absence of facial feature information. The method comprises the following steps: (1) labeling face images at times of rotation of 0 degree, plus and minus 15 degrees, plus and minus 30 degrees and plus and minus 45 degrees to establish a database; (2) using angle labels to carry out feature extraction through five layers of a convolution module, transforming features into 2048-dimensional feature vectors through two layers of full connection, using the feature vectors as inputs of a softmax classification layer, carrying out classification, and thus establishing a convolutional neural network structure; and (3) based on an ubuntu 16.04 operating system, training the convolutional neural network under a GPU of 1080 and a deep learning framework of a CAFFE to obtain a face angle recognition model. The method is used for deep learning-based real-time recognition of three-dimensional face angles.

The invention provides a gesture pre-judging method and system. The method includes: detecting the current gesture initiating coordinates of a gesture guider, and sequentially recording the sequence of the current motion trajectory coordinates, starting from the initiating coordinates, of the gesture guider; extracting the current motion trajectory feature vectors of the gesture guider according to the sequence at every other first preset time interval; using a pre-trained gesture recognition model to perform mode matching on the extracted current motion trajectory feature vectors, and outputting matched complete gesture identification; wherein the pre-trained gesture recognition model is acquired by training pre-collected gesture feature vector sample sets, and the pre-collected gesture feature vector sample sets comprise a plurality of partial motion trajectory feature vectors formed by cutting off continuous features from the beginning of the complete motion trajectory feature vectors of complete gestures according to preset proportions. By the gesture pre-judging method and system, real-time gesture recognition can be achieved.

The invention relates to a method for real-time recognition of USB (Universal Serial Bus) device hot plugging, in particular to a method for real-time recognition of USB hot plugging under a Linux system. The method provided by the invention comprises three function parts, which are detecting the plugging and the unplugging of a device in a real-time manner, confirming the fact that the plugged or unplugged device is a USB device, and confirming the device name, the manufacturer information and the device node. Compared with the prior art, the method provided by the invention has the following advantages: (1) real-time recognition of USB device hot plugging is realized, and sufficient device information is acquired; (2) and a data collecting program compatible with a plurality of devices in accordance with the actual conditions of a plurality of healthy devices is developed; and the total technology process is simple, the recognition principle is reliable, the information data collecting is accurate, and the processing effect is sophisticated, so that the method can be applied to the USB hot plugging situations of various electronic devices and TV digitalization devices.

The invention provides a monitoring method and device based on human body characteristics, and the method comprises the steps: obtaining the human body characteristics of a target object, and receiving a video image which is uploaded by monitoring equipment, screened out from a monitoring video, and provided with a monitoring object; extracting human body characteristics of at least one monitoredobject contained in the video image; calculating a first feature similarity between the human body feature of the target object and the human body feature of each monitoring object;, determining the monitoring object with the first feature similarity greater than the first preset similarity as a target monitoring object, and sending warning information for the target monitoring object, the warninginformation comprising uploading identification information of a video image monitoring device comprising the target monitoring object. Through the method, the monitoring efficiency can be improved.

The invention relates to an expression recognition method and device. The expression recognition method comprises the steps: obtaining a feature area of at least one facial organ in a to-be-recognizedface image, and obtaining at least one feature area image; inputting the at least one feature area image into a corresponding trained expression recognition neural network, and outputting a corresponding feature vector by the expression recognition neural network to obtain at least one feature vector of a first dimension; combining the feature vectors of the at least one first dimension to obtaina feature vector of a second dimension, wherein the second dimension is equal to the sum of the at least one first dimension; and inputting the feature vector of the second dimension into a trained classifier, so that the classifier outputs the expression class information of the face image. According to the embodiment of the invention, the expression recognition speed can be improved while the expression recognition accuracy is ensured, so that the realization of real-time recognition of the facial expression is facilitated.

The invention provides a laser particle image speed measuring data processing method based on moving boundary self-recognition, and belongs to the technical field of hydromechanics experiments. A PIVimage containing laser particles is shot by a PIV camera, image gray values are acquired, and a data matrix is established by the gray values of the PIV image; boundary pixel points in the PIV image are judged and recognized according to a boundary judgment criterion, the boundary pixel points are connected, and a boundary of a measured object in a moving state is acquired; a new SN coordinate system is established according to the boundary and the normal perpendicular to the boundary; the PIV image is subjected to coordinate conversion, data obtained after coordinate transformation of the PIVimage is acquired, and a capture window when the data obtained after coordinate transformation is subjected to correlation calculation always clings to the moving boundary, then a speed vector is acquired by the data obtained after coordinate transformation through a standard correlation calculation algorithm, the speed vector is subjected to coordinate inverse transformation, and a speed flow field of the measured object is acquired. The measuring precision of the speed flow field of the moving boundary is improved.

The present invention provides a human behavior real-time recognition method, and the method comprises the steps of: firstly, collecting triaxial acceleration data of an acceleration sensor by utilizing mobile equipment and uploading the triaxial acceleration to a PC (personal computer) machine; secondly, preprocessing the triaxial acceleration data by utilizing a preprocessing module on the PC machine, wherein the preprocessing includes window length cutting and feature extracting; then training a Gaussian processes classifier model with relatively high precision by utilizing the pre-processed data; transplanting and installing the Gaussian processes classifier model and the preprocessing module to an Android platform; and then installing the platform to the mobile equipment; and finally performing real-time testing and observation on human behaviors. The method can be effectively applied to various actual environments such as emergency rescue environments, medical healthcare environments and the like; by solving the congestion problem caused by too large data volume during remote data transmission in a conventional method, the data volume processed by a mobile phone is reduced; the energy consumption is lowered; and efficient, accurate and real-time remote human behavior monitoring is realized.

The invention provides a small and medium-sized navigation channel bridge monitoring device characterized in that the device is arranged on a small and medium-sized bridge and directly facing a navigation channel so as to monitor a ship target in the navigation channel, is arranged on the bridge and comprises a camera which is arranged on the small and medium-sized bridge, directly faces the center line of the navigation channel and is used for shooting the navigation channel so as to obtain a corresponding real-time monitoring video; a radar which is arranged on small and medium-sized bridgesand is at least used for continuously scanning the navigation channel and obtaining a real-time scanning position when a ship target is scanned; a warning module which is used for sending an acousto-optic warning to a ship target in the navigation channel; and a central processing unit in communication connection with the camera, the radar and the warning module, wherein the central processing unit comprises a neural network model storage part, a monitoring information acquisition cache part, a ship target identification part, a video coordinate conversion part, a radar coordinate conversionpart, a data fusion part, a track information identification part, a track prediction part and an early warning analysis output part.

The invention provides a student classroom learning state online evaluation method and system. The student classroom learning state online evaluation method comprises the following steps: synchronously collecting video images and sound signals of students; correspondingly inputting the video image and the sound signal into the trained expression recognition model and the trained emotion state recognition model, and outputting an expression category, an emotion state category and a corresponding probability corresponding to the student, wherein the categories of the expressions, the emotional states and the classroom learning states are the same; and inputting the expression recognition result and the emotional state recognition result into a student state evaluation model, and outputting an evaluated student classroom learning state category.

The invention relates to the field of big data and discloses a black product gang identification method, device and equipment and a storage medium. The method comprises the following steps of within apreset time range, after receiving request information sent by the same user account each time, extracting a plurality of node features in the request information; based on each node feature, constructing a risk feature graph of the user account, and segmenting the risk feature graph into a plurality of feature sub-graphs; respectively calculating the shortest coding length of each feature sub-graph by adopting a preset coding measurement algorithm, and respectively splitting each feature sub-graph into a plurality of feature communities according to the shortest coding length; and inputtingeach characteristic community into a preset black product gang identification model for identification, and outputting a black product gang identification result. The invention also relates to a blockchain technology, and the request information is stored in the blockchain. According to the method, real-time identification of the black gang is realized, and the defense capability of the black gangis improved.

The invention discloses a civil aircraft test flight vibration fault prediction model which is established through the following steps: 1, establishing a database which is composed of a characteristicparameter set and states corresponding to characteristics; wherein the characteristic parameter set comprises time domain characteristics extracted from the vibration parameters of each part of the civil aircraft, frequency domain characteristics extracted from the vibration parameters of each part of the civil aircraft, and flight parameters of which the correlation with the vibration parametersis greater than 0.6; wherein the time domain characteristics comprise the maximum value, the minimum value, the peak-to-peak value and the amplitude change rate of the vibration amplitude; wherein the frequency domain characteristics comprise frequency composition, frequency size and a change trend of the frequency along with time; and 2, establishing a civil aircraft test flight vibration faultprediction model by using the binary support vector machine model to obtain the binary support vector machine model so as to determine the early warning boundary of the test flight data, and taking the early warning boundary as the basis for judging the normal and abnormal single states of the data. The invention also discloses a civil aircraft test flight vibration fault prediction system.

The invention relates to a method and system for identifying a salient-point mould number of a revolution-solid glass bottle based on images. The method comprises the following steps of: 1) rotating the to-be-identified glass bottle on a turn table, and triggering a camera to collect a panoramic image containing a pitting mould number area by virtue of a photoelectric sensor and a rotary encoder;2) preprocessing the collected image to improve the quality of the image; 3) on the basis of smoothening the image rapidly by virtue of an integral image, segmenting the image by virtue of a dynamic threshold, and carrying out morphologic processing on the image to improve the segmentation effect, screening salient points of a mould according to the features (such as shapes, area and coordinates and the like) of the salient points of the mould, and extracting the coordinates of each salient point; 4) according to the distance relationships among the salient points, calculating corresponding symbolic expressions; and 5) inquiring a glass bottle mould code table, and determining the mould number according to the symbolic expressions obtained in the step 4). According to the invention, lightpaths are transformed by a mirror, thereby satisfying the needs on the space layout and adjustment of equipment; the collection of the high-definition salient-point images of the mould is ensured; and the identification system is simple in structure, low in cost, high in precision, high in speed, and easy to popularize and use.

The invention relates to a road water accumulation point identification system, which comprises a vehicle speed evaluation value setting unit, a vehicle flow evaluation value setting unit, a rainfallwater accumulation probability correction coefficient setting unit and a water accumulation identification unit, wherein the vehicle speed evaluation value setting unit sets the vehicle speed evaluation value of a detected road section according to the current average vehicle speed of floating vehicles on the detected road section and a road section water accumulation vehicle speed threshold; thevehicle flow evaluation value setting unit sets the vehicle flow evaluation value of the detected road section according to the current floating vehicle flow on the detected road section and a road section water accumulation floating vehicle flow threshold; the rainfall water accumulation probability correction coefficient setting unit sets a rainfall water accumulation probability correction coefficient according to the current hourly rainfall on the detected road section and multiple rainfall thresholds; and the water accumulation identification unit enables the vehicle speed evaluation value of the detected road section, the vehicle flow evaluation value and the rainfall water accumulation probability correction coefficient to be multiplied to obtain a water accumulation integrated evaluation value of the detected road section, and according to the water accumulation integrated evaluation value, whether the detected road section is a water accumulation road section is judged. Detection of water accumulation without device can be realized, and the device layout and maintenance cost is greatly reduced.

The invention relates to a video image recognition system and method based on deep learning, and the method comprises the following steps: collecting video information and first picture information, decomposing the video information into a plurality of continuous single-frame pictures, and obtaining second picture information; inputting the first picture information and / or the second picture information into a clustering model for clustering classification; determining a clustering center of each type of attitude, and dividing each type of attitude sample into subsets; optimizing a neural network model by using a training strategy of course learning according to the divided subsets; and receiving to-be-identified picture information, and performing attitude identification by using the optimized neural network model. Compared with the prior art, the invention has the advantages that more effective features with more distinguishing strength can be extracted, so that the trained classifier has better generalization performance, the processing speed is high, the posture classification and recognition effect is improved, and the picture recognition accuracy is high.

The invention discloses a low-cost tomato leaf disease identification method based on a lightweight deep neural network, and the method comprises the following steps: collecting a tomato leaf image data set, expanding the collected tomato leaf image data set through a data set expansion method to obtain an expanded image database, and carrying out the preprocessing of the image data set; constructing an improved residual neural network recognition model, and inputting the preprocessed image data set into the improved residual neural network recognition model to complete the training of the model; and recognizing a to-be-detected picture to be actually detected by using the trained model. According to the invention, an improved residual neural network identification model is adopted; disease identification is carried out on tomato leaves through cooperation of separable multi-scale convolution module1 and module2. According to the method, the network width is expanded, the accuracy reaches a relatively high level, the occupied memory is relatively small, real-time identification of tomato diseases on a low-performance terminal can be realized, and the method can be popularized to crop disease identification of other similar application scenes.

The invention provides a distraction driving detection method based on face posture estimation and sight deviation. The distraction driving detection method comprises the following steps: a face detection step, a face tracking step, a face key point detection step, a face pose estimation step and a distraction driving detection step. According to the invention, a depth infrared camera and other sensor equipment are not needed, and only RGB three-channel color images collected by a common camera are needed to be used for vehicle-mounted video image identification and distracted driving detection.

The embodiment of the invention provides a tire wear degree identification method and device. The method comprises the steps that the circumferential acceleration sequence and the radial accelerationsequence of a tire in the driving process are acquired, and the sequences are normalized according to the vehicle speed square of the same moment; the normalized circumferential acceleration sequenceand radial acceleration sequence form a sample to be predicted in the form of a two-dimensional matrix; and the sample to be predicted is inputted to a preset convolutional neural network model, and the wear degree of the tire is determined according to the output result of the convolutional neural network model. According to the method, modeling of the complex mechanism of tire wear can be avoided, and wear degree identification can be realized by analyzing the circumferential and radial acceleration response of the tire so that the method is simpler and more convenient and higher in precision and online real-time prediction can be realized. End-to-end wear degree prediction is performed on the recombined acceleration signal matrix by using the powerful feature extraction function of theconvolutional neural network without too much early pretreatment so as to have robustness for the change of the driving condition.

The embodiment of the invention discloses a behavior recognition method. The method comprises the following steps: extracting skeleton key points of each to-be-detected object in a current frame image; according to the skeleton key points, searching target objects with arm included angles smaller than an included angle threshold value from all the to-be-detected objects; acquiring a hand region image of the target object; and identifying whether the hand region image of each target object contains the target object or not by using a pre-trained behavior identification model, and outputting a behavior identification result corresponding to each target object in the current frame image. Through combination of skeleton key point analysis and hand region image target identification, real-time identification of smoking and mobile phone playing behaviors of people in a monitoring image can be realized. Cigarette mobile phone recognition is carried out on the hand area image, background interference can be greatly reduced, and the recognition accuracy is improved.

The invention discloses an intelligent human-computer interaction method based on dynamic gesture recognition, belonging to the technical field of human-computer interaction. The invention generally adopts the traditional machine learning algorithm for the current gesture-based human-computer interaction scheme, and cannot meet the requirements of real-time and robustness of the embedded equipmentunder the background of a smart home. By improving the lightweight target detection network, the invention realizes the fast and accurate detection of the hand region, on the basis of which the target tracking algorithm is integrated to obtain the motion trajectory of the hand and the personalized human-computer interaction behavior is provided according to the trajectory classification result. The invention can realize the real-time recognition of the dynamic gesture on the embedded device and has strong robustness to factors such as illumination, skin color, background and the like, and isan intelligent human-computer interaction solution facing the intelligent family scene.

The invention relates to a multi-scene smoke detector based on bidirectional blue light detection and a self-adaptive identification method, which belong to the technical field of fire-fighting fire alarm systems. The method comprises the following steps: S1 a fire detector converts the detected aerosol concentration into an electric signal; S2 amplification filtering and A / D conversion are carried out on the voltage signal through a signal processing circuit; and S3 a digital signal obtained through A / D conversion is analyzed in a microprocessor, signal feature extraction, analysis and identification are achieved, a scene is finally judged, and whether an alarm needs to be given or not is judged. The multi-scene smoke detector comprises two blue light emitting diodes and a blue light receiving tube, and the two blue light emitting diodes and the blue light receiving tube form 70 degrees and 135 degrees in space respectively to form forward scattering and backward scattering; real-time identification, scene judgment and threshold adjustment of fire data can be realized through a bidirectional blue light technology, and a foundation is laid for practical application of fire detection.

The invention provides an empty parking space identification method. The empty parking space identification method comprises the following steps: 1, acquiring image information of a parking space by using a camera mounted above the parking space; 2, for the image information, determining a first confidence coefficient that the parking space is an empty parking space by using a first identificationalgorithm; 3, for the image information, determining a second confidence coefficient that the parking space is an empty parking space by using a second identification algorithm; 4, judging whether the parking space is an empty parking space or not based on the first confidence coefficient and the second confidence coefficient.

The invention discloses a method for recognizing dangerous driving behaviors of a driver based on intelligent equipment. The method is mainly used for the dangerous driving behaviors including sharp turning of a steering wheel, frequent eye rubbing and using a mobile phone. According to the method, data of an acceleration sensor, a gravity sensor and a gyro sensor in the intelligent equipment is utilized to realize recognition of the dangerous driving behaviors of the driver so as to improve the practicability and accuracy of recognition, no interference with normal driving of the driver is caused, and practicability of recognition of the dangerous driving behaviors of the driver can be effectively improved. Meanwhile, the dangerous driving behaviors of the driver can be recognized in realtime, when the driver generates the dangerous driving behaviors, an early warning prompt is performed, and occurrence of traffic accidents caused by the dangerous driving behaviors can be effectivelyavoided.