32results about How to "Avoid low recognition accuracy" patented technology

The invention discloses an intelligent flow control system of an interactive network, comprising client-sides, a router and an intelligent flow control module, wherein the router is communicated with the client-sides by networks, and the intelligent flow control module is arranged on the router. The intelligent flow control system is characterized in that each client-side is provided with an application identification module for carrying out application type identification on the data streams of the client-side. The invention also discloses an implementing method of the intelligent flow control system of the interactive network, comprising the steps as follows: (a) the application identification modules are used for identifying the data streams generated by the client-sides, and transmitting the data streams to the router after program identification codes are stamped on the data streams; and (b) a classification module on the router receives the data streams from each client-side and stores the data streams in a classification mode. Compared with the prior art, by using the system and the method provided by the invention, the identification accuracy of the router for a personal computer (PC) is greatly improved, the problems of network blockage, disconnection, system crash and the like caused by overload of the router can be avoided, and the practical value is very high.

The invention discloses a micro-expression recognition method based on deep learning. The micro-expression recognition method comprises the following steps: 1, cutting, framing and extracting video data containing expression actions; 2, performing face alignment, face clipping, normalization and other preprocessing on the extracted expression sequence; 3, performing data enhancement operation on the obtained data set; 4, establishing a neural network model; 5, dividing all facial expression data into a training set and a test set in proportion; and 6, testing the model by using the test set, outputting information such as identification accuracy, identification time, errors and the like, and selecting the current model when the identification rate meets the requirements. The micro-expression recognition method based on deep learning has the advantages of being simple, efficient, low in cost, high in precision and the like.

The invention discloses a mine slight shock and blasting signal identification method based on waveform slope before and after a peak value. The method comprises: step 1, acquiring a linear identification equation: based on N times of slight shock events and M times of blasting events, obtaining a linear identification equation with logarithm of A, logarithm of k1, logarithm of k2, logarithm of k3, logarithm of k4, logarithm of k5, and logarithm of k6 as characteristic parameters, the linear identification equation being Y=a1* lg(k1)+a2*lg(k2)+a3*lg(k3)+a4*lg(k4)+a5*lg(k5)+a6*lg(k6)+b*lg(A)+c; step 2, calculating a distinguishing threshold value Yd; step 3, based on the linear identification equation and the distinguishing threshold value Yd, identifying a to-be-identified event; and calculating a distinguishing value Y of the to-be-identified event, if Y<=Yd, the to-be-identified event being a slight shock event, and if not, the to-be-identified event being a blasting event. The method is low in calculated amount, and identification results are not influenced by signal noise, and the method is high in accuracy.

The invention relates to an ocean front area acquisition method and device, computer equipment and a storage medium. The method comprises the following steps: determining a roughly estimated ocean front area, target paths for a plurality of unmanned naval vessels to reach the roughly estimated ocean front area and target duration for the unmanned naval vessels to reach the roughly estimated oceanfront area along the target paths according to acquired satellite remote sensing observation data; determining each initial position of each unmanned ship in the roughly estimated ocean front area according to received arrival confirmation information sent when each unmanned ship reaches the roughly estimated ocean front area; outputting an adjustment instruction according to the region change information of the roughly estimated ocean front region determined by the multiple groups of satellite remote sensing observation data received within the target duration to indicate each unmanned ship to perform adjustment based on the respective corresponding initial position so as to obtain the adjusted position of each unmanned ship; and determining a target ocean front area according to each received adjusted position. By adopting the method, the identification accuracy of the ocean front area in the sea area can be improved.

The invention discloses a trip mode identification method based on GPS trajectory data. For trip modes lack of a GPS signal and having an incomplete signal, a subway identification method based on a rule is adopted; and, for other trip modes having normal GPS signal records, a random forest classifier is adopted, and modeling and verification are performed through a combination with the GPS trajectory data. Before trip modes are identified, 4 kinds of characteristic parameters related to mode identification are screened through a characteristic selection process, and all the real trip modes can be accurately identified. The method is not limited by the GPS data and not relied on other data sources, and the method has the advantages of high universality, simple calculation and accurate and reliable identification results; and the method is suitable for trip mode identification based on the GPS trajectory data, and is in favor of large-scale promotion of trip survey based on intelligent mobile phones.

The invention provides a method for detecting the bad appearance of a product, and the method comprises the following steps: obtaining a product appearance image, carrying out the graying of the product appearance image, and obtaining a first gray level image; binarizing the first grayscale image to obtain a second grayscale image; performing contour extraction on the second grayscale image, and screening to obtain a contour combination; obtaining a minimum bounding rectangle of the contour combination, completing appearance feature classification of the product appearance image through a neural network according to the minimum bounding rectangle, and identifying a product with poor appearance according to a classification result. According to the invention, contour extraction is carried out by sequentially carrying out graying and binaryzation, so that the reduction of the recognition accuracy caused by the influence of light is avoided; and by combining the minimum bounding rectangleof the contour combination with the neural network, the occupation of operation resources is reduced, the image processing speed is at a millisecond level, the operation real-time performance is relatively high, the detection effect is relatively good, automatic identification and testing can be realized, and the invention can be widely applied to the technical field of product quality detection.

A method and an apparatus for character string recognition may be provided that enables prevention of a decrease in recognition accuracy for a character string even when distortion of an image appears in a direction perpendicular to a medium transfer direction.

The invention discloses a recognition model training method, device and equipment and a storage medium, and relates to the field of financial science and technology, and the method comprises the following steps: obtaining a to-be-trained image, and constructing a simulation image according to the to-be-trained image; determining a training data set according to the to-be-trained image and the simulation image; and based on the training data set, training through a generative adversarial network and a recognition network in a preset neural network model to obtain a recognition model. Accordingto the method, the training data set is obtained by constructing the simulation image through the obtained image, and the situation that the recognition accuracy of the recognition model obtained through training is low due to the fact that samples of the training data set are insufficient is avoided, that is, the recognition accuracy of the recognition model obtained through training is improved.

The invention provides a driver behavior recognition method based on multi-source information fusion. The driver behavior recognition method is characterized by comprising the steps that 1, images or videos related to driver behaviors recorded in a driving simulator and vehicle motion state data are obtained; step 2, inputting the processed driver behavior data into a fine-tuned Vision Transform first sub-model, and outputting a probability matrix P1 of four driving behavior categories; 3, preprocessing the synchronously recorded vehicle motion state data, inputting the data into the trained Bi-LSTM second sub-model, and outputting a probability matrix P2 of four driving behavior categories; and step 4, calculating the probability output matrixes P1 and P2 obtained in the step 2 and the step 3 to obtain information entropies H1 and H2 of the first sub-model and the second sub-model, and finally performing weighted decision fusion on Softmax function output probability values of different sub-models to realize final classification of the four driving behaviors. According to the invention, a plurality of driving behavior identification tasks are integrated, and the anti-interference performance is high.

The invention discloses a ship trajectory classification method based on feature selection and hyper-parameter optimization. For the problem of unbalanced original data in ship trajectory classification, firstly, data cleaning and preprocessing are performed on multiple pieces of trajectory data of an automatic ship identification system, then all trajectories are drawn by matplotlib, unavailable trajectories are deleted, then features are calculated to obtain additional features, dimensionality reduction processing is performed on speed, course and trajectory coordinates, backward feature selection is carried out on all features after dimension reduction processing through a random forest (RF), finally hyper-parameter optimization is carried out through the random forest, and classification prediction is carried out on the features through model training and performance evaluation without depending on an external information source. The ship trajectory classification method has high performance and is stable, and can be effectively applied to actual ship trajectory classification.

The invention provides a dialogue intention recognition method and device, electronic equipment and a readable storage medium. A natural language understanding model is utilized to determine a possibility score of each intention corresponding to a consultation dialogue; if a real intention capable of representing the consultation dialogue does not exist in at least one first intention determined according to the possibility score, determining a target keyword from the consultation dialogue by utilizing a configured business keyword model, and searching at least one second intention matched with the target keyword in a business knowledge graph; and if the at least one second intention does not include the first intention with the highest possibility score, further determining a real intention indicated by the consultation dialogue from the at least one second intention according to the feedback of the user. In this way, the accuracy of user intention recognition can be improved, and the situation that due to the fact that the character generalization ability of a natural language understanding model is insufficient, the recognition accuracy is low, and the intention of the user cannot be recognized is avoided.

The invention discloses an accessory device for array particle collision sensors. An impurity discharging mouth is formed in the tail of a combine harvester, a mounting rack is sequentially provided with a screen connecting bracket, a funnel connecting bracket, a sensor accessory mounting bracket and a sensor mounting bracket from top to bottom, and the bottom of the mounting rack is hinged to thebottom of the impurity discharging mouth; the angle of the mounting rack can be adjusted and fixed, the positions of a screen and a funnel relative to the impurity discharging mouth can be adjusted,and the sensor accessory mounting bracket is installed on the mounting rack; and the sensor accessory is adjustable, and is vertically corresponding to the outlet of the lower end of the funnel, the sensor mounting bracket is mounted on the mounting rack, and the array particle collision sensors are located directly below the sensor accessory. Through the accessory device, the angle, speed and position of material collision on the particle collision sensors can be adjusted, the problem of low recognition accuracy rate which is caused by random speed and angle of particle collision can be avoided, and meanwhile, occurrence of particle collision in a gap of detection units of the array particle collision sensors can be avoided, so that the detection accuracy of the particle collision sensorsis improved greatly.

The invention discloses a license plate recognition method, device, and equipment and a medium. The method comprises the steps: acquiring a to-be-recognized license plate image; performing gray processing on the to-be-recognized license plate image to obtain a to-be-recognized gray license plate image; and inputting the to-be-recognized gray license plate image into a license plate recognition model for license plate recognition to obtain a license plate number sequence corresponding to the to-be-recognized gray license plate image. According to a traditional license plate recognition method, characters in a license plate image are segmented by adopting an artificial design algorithm in combination with projection, connection and/or contour extraction and other methods, a single character image is obtained, and then character recognition is performed on the single character image through a classifier, and thus technical problems of low recognition precision and low recognition speed of a subsequent character recognition result caused by the fact that character segmentation quality is easily influenced by factors such as noise, low resolution, blurring or deformation of an input image exist in the prior art are caused. However, with the provided method, the problems in the prior art are solved.

The invention discloses a bridge floor multi-axis moving load recognition method based on the diagonal relaxation orthogonal projection iterative algorithm. The method includes the following steps that firstly, m displacement sensors are attached to corresponding positions x1, x2,..., xm on the bottom face of a bridge, and the displacement of a bridge floor multi-axis moving vehicle load fk(t) at the x position at the t moment is v(x,t), wherein k is 1,2,3,..., and k is the number of vehicle axles; secondly, an oscillatory differential equation is established; thirdly, the equation (1) is solved; fourthly, a multi-axis moving load system equation recognized by displacement responses is established for the bridge under the k-axis vehicle load action; fifthly, the precise value of a multi-axis moving load is solved through the diagonal relaxation orthogonal projection iterative algorithm. The multi-axis moving load can be recognized by measuring the bridge displacement responses only, the recognition method is simple, precision is high, feasibility is good, and the method can be widely applied to moving load recognition of various types of bridges.

An image processing apparatus (12) has a processing device (122) for executing, on a first image (111w) captured by a first imaging device (11w) imaging a first imaging area and a second image (111t) captured by a second imaging device (11t) imaging a second imaging area, an image recognition processing for recognizing a target in the first image and the second image, the processing device executes the image recognition processing on a non-overlapped image part (113w) of the first image without executing it on a overlapped image part (112w) of the first image when the processing device executes the image recognition processing on the first image, the non-overlapped image part includes a non-overlapped area at which the first imaging area does not overlap with the second imaging area, the overlapped image part includes a overlapped area at which the first imaging area overlaps with the second imaging area.

The invention discloses an intelligent lock fingerprint recognition device capable of prolonging the service life. The fingerprint recognition device comprises a fingerprint recognition module and a glass panel, wherein the fingerprint recognition module is provided with a fingerprint acquisition module for reading the resistance change of the skin of a finger of a user. The fingerprint acquisition module is provided with an acquisition panel, and the acquisition panel clings to the bottom surface of the glass panel, wherein the surface of the glass panel is provided with a fingerprint acquisition area for the finger of the user to touch and sense, and the fingerprint acquisition area is opposite to the fingerprint acquisition module. The device can reduce the loss, and the service life is remarkably prolonged.

The invention relates to the field of data processing, and provides a method for identifying a table in a PDF document, which comprises the following steps: acquiring an original target page of the PDF document; when it is recognized that no conventional table exists in the original target page, or when the recognition result of the conventional table in the original target page does not meet a preset condition, recognizing whether a special table exists in the original target page or not through a special table recognition module, and the conventional table referring to a table with crossed lines; the special table being a table without cross lines, and the special table identification module is a module for identifying the table without cross lines. According to the method, the identification accuracy of the table in the PDF document can be improved.