65results about How to "Avoid recognition errors" patented technology

According to the invention, this air conditioner includes an air regulating portion that absorbs indoor air and discharges the air to the outdoor after regulation; an infrared sensor module disposed above said air regulating portion; a control portion that controls said air regulating portion based on a detection result from said infrared sensor module, and includes a substrate; an infrared sensor disposed on said substrate; a shell combined with said substrate; and a shell rotary mechanism that rotates said shell. According to this air conditioner provided in this invention, it is provided with high precision for human body detection, also, the infrared sensor, together with the substrate, is integrated with the shell; therefore, the air conditioner is impact and simplified.

The embodiment of the invention provides a method and device for recognizing images. The method and device for recognizing the images are used for solving the problem that optical character recognition errors are caused by poor image quality. The method comprises the steps of obtaining the images to be recognized, detecting the boundary outline of each character from the images to be recognized, searching for the outline pixel points on the boundary outline of each character, and detecting the pixel difference values of the outline pixel points, carrying out statistics on the pixel difference values of the outline pixel points on the boundary outline of each character in the images to be recognized, judging the definition of the images to be recognized according to the statistic result of the pixel difference values, and carrying out optical character recognition on the images to be recognized after the images to be recognized are judged to be clear. Fuzzy images to be recognized are excluded before optical character recognition is carried out, subsequent optical character recognition can be carried out on the clear images to be recognized, the problem that recognition results are wrong due to the low definition of the images is avoided, and the recognition efficiency is improved.

The invention discloses a speech-emotion recognition method based on improved Fukunage-koontz transformation (FKT). Transformation by utilizing the method can effectively realize feature dimension reduction. During dimension reduction, in consideration of the internal manifold structure of speech, a parameter tau capable of describing inter-frame relevance is introduced, and weighting is performed when feature covariance is calculated so as to find the feature with maximum/minimum variance after dimension reduction. In addition, by taking the variance as the discrimination information of classification, the recognition of various speech emotions is realized by adopting a k nearest neighbor recognition method. Compared with the prior similar recognition method, the method can effectively improve recognition rate.

The invention provides an intelligent security check method based on terahertz imaging. The method comprises the following steps: obtaining a terahertz image of a detected object from a terahertz imaging system, and preprocessing the terahertz image to obtain a preprocessed terahertz image; inputting the preprocessed terahertz image into a preset deep learning model, and extracting object information data of a corresponding object to be detected in the preprocessed terahertz image; carrying out image instance segmentation and detection identification on the received article information data byusing a deep learning model; and fusing the instance segmentation result and the target detection and identification result, and generating a corresponding identification result according to the information judged as the suspicious prohibited article. According to the technical scheme provided by the embodiment of the invention, instance segmentation and detection identification of the target object in a variable environment are realized, the identification accuracy of prohibited goods in the security inspection process is ensured, and the complete information of the target object can be obtained.

The invention discloses a current and voltage state monitoring method and device based on fundamental wave zero sequence characteristics. The current and voltage state monitoring method comprises the following steps: obtaining three input variables, that is, a three-phase voltage fundamental wave zero sequence value, a three-phase current fundamental wave zero sequence value, and a zero-sequence current and neutral current difference absolute value; inputting the three input variables into a preset fuzzy inference system, wherein the fuzzy inference system comprises a mapping relation between the three input variables and fault type diagnosis results serving as output variables; and finally, obtaining the fault type diagnosis results. The current and voltage state monitoring device comprises an input variable obtaining program unit and a fuzzy inference program unit in full correspondence to the steps of the method. The current and voltage state monitoring method and device based on the fundamental wave zero sequence characteristics can accurately judge various faults under different conditions, are high in detection accuracy, do not cause false alarm, and can realize fault alarm and effective fault removal, thereby ensuring safety and stability of the power supply system.

The invention provides an improved mobile phone signaling data trip identification method based on a dynamic space threshold, which is used for identifying travel conditions of each user according tomobile phone signaling data acquired by each mobile phone base station so as to obtain an individual travel chain of the user, and is characterized by comprising the following steps: S1, acquiring themobile phone signaling data; s2, cleaning the mobile phone signaling data; s3, judging a dynamic threshold D corresponding to each mobile phone base station according to the position of the base station through a preset dynamic threshold calculation method; s4, performing classifying to obtain a to-be-analyzed data set of the user; s5, sequentially selecting a group of to-be-analyzed data sets ofthe user; s6, reading to-be-analyzed data in the current user data set in sequence, judging the stay point of the current user according to the corresponding dynamic threshold value D, S7, generatingand storing an individual travel chain, and S8, repeating the step S5 to the step S7 until all the to-be-analyzed data of the user are processed.

An embodiment of the invention relates to the field of information technology processing, in particular to an associated address acquisition method, an associated address acquisition device, electronic equipment and a storage medium. The associated address acquisition method comprises the following steps of: receiving a message or an instruction from a client; analyzing the message or the instruction through at least one processor to obtain a target address, wherein the target address comprises longitude and latitude information and text description; determining a spatial index code of the target address through at least one processor according to the longitude and latitude information; and determining an interest surface of the target address through the at least one processor according to the text description, wherein the associated address of the target address comprises the spatial index code and the interest surface. By adopting the associated address acquisition method and the associated address acquisition device of the invention, a more accurate associated address can be obtained.

The embodiment of the invention provides a method and device for identifying the number of to-be-detected cargoes. The method comprises the steps of collecting a front image and a top image of a wholesupport of to-be-detected cargoes; performing white balance processing on the image; inputting the white balance front image into a first segmentation model to obtain a segmented front container image; inputting the white balance top image into a second segmentation model to obtain a segmented top container image and the number of vacant layers corresponding to the top container image; and basedon the segmented front container image, the segmented top container image and the number of vacant layers corresponding to the top container image, determining the number of to-be-identified cargos ofthe whole support; wherein the first segmentation model and the second segmentation model are obtained based on machine learning. According to the method and the device provided by the embodiment ofthe invention, the accuracy and the efficiency of cargo quantity statistics are improved.

The present invention discloses a method for identifying personal emotional arousal degree using human electrocardiogram (EEG), skin electrocardiogram (EDA), sound signal and facial video signal. TheECG, skin electrogram, voice and facial video signals were collected and 22 features are extracted as training data. Emotional arousal level is evaluated by professionals as evaluation data. Feature extraction and feature data standardization are carried out on the training data in turn. Support vector regression (SVR) model is trained by using the processed training data and evaluation data. Whenpredicting and identifying the emotional arousal, the signal data is input to the trained model, and the model is used to predict and identify the emotional arousal state level of the individual.

The invention discloses a live broadcast content risk information control method and system. Capturing risk content in the live channel through a risk information capturing system and generating a first work order; then, primary manual auditing is performed in the first auditing system; obtaining a first audit result, if the first auditing result is suspected violation, the first auditing result is not suspected; if yes, generating a second work order and sending the second work order to a second auditing system; secondary manual rechecking is carried out in the second auditing system in combination with the real-time audio of the live channel; according to the invention, the identification error of the audio identification technology is avoided, and the technical problems that the existing risk information control system is single in identification technology and large in error, the audio identification technology enables the enterprise cost to be improved, and accurate, rapid and efficient risk information control cannot be realized are solved.

The invention relates to a Faster R-CNN-based railway bullet train hood front opening and closing damage fault identification method. The invention relates to the technical field of headstock cover image recognition. The method comprises the steps of collecting an original image of a headstock cover; preprocessing the collected original image to reduce image noise; performing deep learning according to the preprocessed image to obtain a training sample; performing deep learning model training according to the obtained training sample to obtain a trained fault identification model; and according to the trained fault identification model, carrying out damage fault identification on the headstock cover. According to the method, the train hood front opening and closing damage fault is identificated and detected through a deep learning method, and identification errors caused by fatigue and personal judgment differences during manual detection are effectively avoided. Compared with manual detection, the deep learning method can improve the detection efficiency.

The invention relates to the technical field of human body recognition, and discloses a posture recognition method, device and equipment based on a 3D human body and a storage medium, and the method comprises the steps: obtaining a to-be-recognized image, and segmenting the to-be-recognized image to obtain a to-be-recognized human body image; acquiring a human body posture deep learning model based on the 3D human body model; and recognizing a to-be-recognized human body image according to the human body posture deep learning model. Deep learning is performed by using the 3D human body model to obtain the human body posture deep learning model, the human body posture can be identified based on the human body posture deep learning model, and the preset human body posture comes from the 3D human body model, so that special posture actions do not need to be obtained through a large number of real person experiment sampling; the recognition error caused by the difference between the experimental sampling image and the to-be-recognized object is avoided, and the human body posture recognition effect is improved.

The invention relates to tea fresh leaf classification treatment equipment and a classification treatment method. The tea fresh leaf classification treatment equipment comprises a reciprocating feed assembly, a vibration platform assembly and a classification platform assembly; and the vibration platform assembly comprises a U-shaped groove vibration platform, and a first driving mechanism drivesa feed hopper to reciprocate to uniformly feed materials into each U-shaped groove of the vibration platform. A second driving mechanism realizes the transverse and longitudinal two-degree-of-freedomvibration of the vibration platform; the classification platform assembly comprises a conveying belt and a plurality of hollow sliding plates which are arranged in a ladder shape and have adjustable inclination angles and are used for allowing fresh tea leaves to slide from top to bottom in sequence, compressed air forms an air cushion on the upper surface of a sliding plate which is full of air holes, an executing component assembly which comprises an upper camera assembly and a lower camera assembly which form an included angle of 90 degrees and a high-speed electromagnetic valve nozzle assembly is arranged between the adjacent sliding plates, the upper camera assembly and the lower camera assembly simultaneously take pictures of the fresh tea leaves passing through the middle area of the adjacent sliding plates, after algorithm identification, and the high-speed electromagnetic valve nozzle assembly blows the target fresh tea leaves onto the corresponding conveying belt to realize classification of the fresh tea leaves.

The invention discloses a tobacco stem and tobacco leaf precise recognition method and system based on dual-energy induction. A dual-energy detection structure is adopted on an X-ray imaging part, and due to the fact that high-energy ray and low-energy ray are different in penetrability on different kinds of matter, corresponding gray values are changed when the energy level is changed. On the basis of the characteristic that gray differences of tobacco leaves and tobacco stems in the same image are large in ray images of different energy levels, the tobacco leaves and the tobacco stems can be accurately differentiated, recognition errors caused by imaging are avoided to the large extent, and tobacco leaf and tobacco stem recognition accuracy is improved; the whole system is simple in structure, only one energy X-ray mechanism needs to be additionally arranged or induction lines of different energy levels are used on the basis of an existing recognition system, cost is low, implementation is easy, operation is convenient, and the method and system have large promotion space.

The invention discloses an entity recognition model training method and device, computer equipment and a storage medium, and the method comprises the steps: firstly obtaining a sample data set, enabling each piece of sample data in the sample data set to comprise N pieces of labeling data, and enabling N to be a positive integer; according to the sample data set, training a preset multi-layer recognition model to obtain an entity recognition model, wherein the multi-layer recognition model comprises a main model and N entity sub-models, and each piece of annotation data of each piece of sampledata corresponds to one entity sub-model. According to the method, a plurality of annotation data are set in the sample data, and when the multi-layer recognition model is trained, the network structure of the main model and the N entity sub-models is set, so that the memory consumption during training can be reduced. Moreover, N pieces of annotation data are set for one piece of sample data, sothat the recognition precision of the model can be better ensured on the premise of not reducing the number of the sample data. The invention further discloses an entity identification method and device, computer equipment and a storage medium.

The invention discloses a data identification method, a junk mailbox identification method and a file identification method. The data identification method comprises the steps that to-be-recognized data is acquired, and the to-be-recognized data comprises character information used for indicating an identifier of the to-be-recognized data; language features corresponding to the character information are obtained, and the language features comprise at least one of the following: the repetition proportion of pronunciation of characters contained in words in the character information, the proportion of repeated characters contained in the words and the proportion of numbers contained in the words; and whether the to-be-identified data is the first type of data based on the language featuresis determined. According to the method, the technical problem of inaccurate identification result caused by a certain identification error in a text similarity identification process when the text similarity is adopted to identify the data in the prior art is solved.

The invention discloses a device and method for detecting film missing defects of resin lenses. The device comprises a dark box (1), a rotary motor (2), a fixing frame (3), a motion controller (4), amaster controller (5), a camera (6), a light source (7), a lens fixing frame (8) and a background plate (9). The device is characterized in that the rotary motor (2) is fixed at the top of the dark box (1), connected with the motion controller (4) and used for rotationally switching positions of the camera (6) and the light source (7) to realize two-time subarea collection of lenses; the motion controller (4) and the master controller (5) are connected through a serial port, and the motion controller (4) is triggered through a serial port signal to drive the rotary motor (2) to execute the rotation action; the fixing frame (3) is connected with the rotary motor (2) through a key slot structure. The device and the method can be applied to detection of film missing defects of various lensessuch as myopic glasses, presbyopic glasses, hyperopia glasses and astigmatoscopes.

The invention provides a radar radiation source signal separation method based on improved variational mode decomposition. The method comprises the following steps: establishing a radar radiation source signal library of multiple modulation modes; constructing a variational model required by a variational mode decomposition algorithm; extracting the Renyi entropy of the additive hybrid radar signal as a fitness value; calculating optimal parameters of the variational mode decomposition algorithm by applying an artificial bee colony algorithm; decomposing the mixed signal into a virtual multi-channel observation signal through variational mode decomposition; signal reconstruction is realized by means of a singular value decomposition and quick independent component analysis method; extracting a time-frequency domain Renyi entropy of the separated signal as a distinguishing feature; and verifying the signal separation effect by using a support vector machine. According to the method, the additive hybrid radar radiation source signals are separated and recognized, the improved variational mode decomposition method is provided for solving the problems that the number of signals detected by a receiver is large, priori information is little and the recognition effect is poor, quick separation and accurate recognition of the hybrid radar signals are achieved, and a brand new thought is provided for follow-up processing of the hybrid signals.

The invention discloses a prediction model construction method, device and system and a bank card number identification method, and belongs to the technical field of computers. The prediction model construction method comprises the following steps: acquiring a bank card sample image of a bank card and calculating the bank card sample image to obtain a candidate area; inputting the bank card sampleimage and the candidate region into a Fast-R-CNN model; and training the Fast-R-CNN model according to a preset classification label, the regression position corresponding to the classification label, a classification prediction result output by the Fast-R-CNN model and a regression position prediction result corresponding to the classification prediction result, and determining the trained Fast-R-CNN model as a prediction model. The method is different from most OCR methods, extra work such as character segmentation and image processing is not needed, the method is an end-to-end method, andrecognition errors caused by image segmentation and other processes can be avoided.

The embodiment of the invention discloses a file index generating method and device. The method comprises the following steps: obtaining an electric file containing a two-dimensional code; distinguishing the two-dimensional code, and obtaining a first character string of professional work information containing the electric file; comparing the professional work information in the first character string with the turnover information of a host; when the professional work information in the first character string and the turnover information of the host are compared to be consonant, establishing an index for the electric file according to the professional work information in the first character string. The embodiment of the invention also accordingly discloses the file index generating device. Through the adoption of the embodiment of the invention, the index of the electric file is generated by using a two-dimensional code technology, the filing process is simplified, and the filing efficiency is improved.

The invention discloses a positioning mechanism of a ship inner cabin rust removal robot capable of accurately identifying a rusty area, which comprises a recovery disc and a frame assembly, and the frame assembly is provided with a transverse beam, two walking assemblies, two clamping assemblies, two supporting assemblies, two extending assemblies and four driving assemblies. The recycling disc is fixedly installed in the middle of the frame assembly, a rusty area positioning component and a high-precision positioning assembly are further arranged in the recycling disc, and the rusty area positioning component determines a rusty area through a comparison algorithm based on images obtained after water drop weakening processing. According to the ship inner cabin rust removal robot capable of precisely recognizing the rusty area, the rusty area can be precisely recognized through the positioning mechanism, turning movement can be achieved through the driving mechanism, and the situation that the rust removal robot is manually carried to adjust the movement direction of the rust removal robot is not needed; the frame is kept at an inclined angle through the supporting assembly and the extending assembly, and adaptive adsorption of magnetic wheels on the frame to corners is achieved.

The invention discloses a railway D-series high-speed train apron board grating deformation fault identification method, solves the problem of low detection efficiency of an existing railway bullet train apron board grating deformation fault identification method, and belongs to the technical field of railway bullet train fault identification. The method comprises the following steps: constructingan apron board grating deformation sample set of the railway bullet train; training a deep learning target detection network Faster R-CNN by using the apron board grating deformation sample set to obtain a Faster R-CNN detection model and a weight; and identifying a to-be-detected side image of the railway bullet train by using the Faster R-CNN detection model and the weight, and determining whether the apron board grating in the side image deforms and the deformation position. The feature extraction network provided by the invention comprises the following steps: replacing a 3 * 3 convolution kernel in a Bottleneck block in a Resnet-50 network with a 3 * 1 convolution kernel connected in series with a 1 * 3 convolution kernel, and replacing a ReLU activation function in the feature extraction network with a Switch activation function. According to the method, the train apron board grating deformation fault is identified and detected, and the identification error caused by fatigue andpersonal evaluation difference during manual detection is effectively avoided.