71results about How to "Meet the requirements of real-time detection" patented technology

The invention provides an electric power equipment infrared image real-time detection and identification method based on artificial intelligence. The method comprises the following steps of S1, acquiring infrared images of various types of electric power equipment through an infrared thermal imager; S2, preprocessing an acquired image to form an effective power equipment infrared image data set; S3, performing target label processing on the obtained data set; dividing the data set into a training set and a test set; S4, constructing an improved YOLOv4 real-time detection model for detecting and identifying an infrared image target of the power equipment; S5, training and parameter adjustment of the model are carried out by using a training set in the data set; and S6, performing target detection and identification on the trained model by using a test set in the data set to prove the effectiveness of the model; through the above steps, automatic detection and identification of infraredimages of various types of power equipment are realized; the accuracy degree of identification can be greatly improved, detection and identification efficiency is improved, and operation resources areeffectively utilized.

The invention discloses a tip timing vibration parameter identification method for a rotating blade. The method comprises the following steps of determining the number of tip timing sensors and circumferential installation positions; collecting an under-sampled signal of a vibration displacement of the rotating blade; building a sparse representation model of the under-sampled signal, and determining a vibration frequency of the blade by using a sparse reconstruction method; and obtaining a vibration order of the blade based on the vibration frequency of the blade determined by the sparse reconstruction method and a rotation frequency of the blade, building a blade vibration equation design matrix based on the vibration order of the blade and the installation positions of the tip timing sensors, and identifying a vibration amplitude of the rotating blade by using a circumferential Fourier method. The invention further provides a tip timing vibration parameter identification system forthe rotating blade. According to the method and the system, no additional prior information is needed; the vibration frequency and amplitude parameters of the blade can be identified by measuring theunder-sampled signal of the vibration displacement only by utilizing the tip timing sensors; the vibration parameter identification step is simplified; and the precision of an identification result ishigh.

The invention provides a detection method for a fatigue state of a driver. Confidence coefficients are acquired through calculation by sending eye SIFT features into a classifying model; the opening-closing states of the eyes are determined according to the confidence coefficients without being affected by a human face angle; and the detection method has a rotation invariable property, high accuracy on eye detection, simple calculation process, and can detect the fatigue state of the driver in time and can satisfy real-time detection requirement. In addition, the invention also provides a detection system for the fatigue state of the driver.

The invention discloses a honeycomb paper core defect detection method based on machine vision. Aims at various defect problems generated during a honeycomb paper core production process, the method comprises the following steps of acquiring a honeycomb paper core picture on a production site; detecting the defects in the honeycomb paper core by adopting an SSD deep neural network; judging the defect category of the honeycomb paper core and outputting the specific position, then using a machine vision algorithm for rapid reinspection to prevent the false inspection, finally transmitting an obtained result to the honeycomb paper core defect repairing system, and providing a correct feedback signal to realize the automatic repairing of the honeycomb paper core defects. The method detects thehoneycomb paper core defects in real time through the deep learning model and the machine vision algorithm, can provide the feedback information for an automatic defect repairing system during the honeycomb paper core production process, has the advantages of being accurate in recognition, accurate in positioning and fast in recognition speed, and can meet the requirement for honeycomb paperboardproduction automation.

The invention belongs to the technical field of video detection in deep learning. The invention relates to a vehicle detection method, in particular to a rapid vehicle detection method based on windowcharacteristics. According to the method, a vehicle window is used for replacing a vehicle body to serve as a target object for detection, a residual module of a ResNet network and a multi-scale feature extraction method of an SSD algorithm are combined, a network structure of YOLOv3 is used for reference, and a full convolution detection method with only 24 convolution layers is constructed; Under the condition that the traffic flow is large, during batch testing, the average detection precision is close to 100%, the average detection rate reaches 90%, the detection speed reaches 22 milliseconds per frame, real-time detection of vehicles in a road high-definition monitoring video is achieved, the detection rate of the vehicles in the large traffic flow is effectively increased, and the method has important application value.

The invention discloses a sow side-lying posture real-time detection system based on joint partitioning of sow key parts and environment, and the system comprises a delivery room, a camera, a video storage unit, and a server, and the delivery room is used for placing a to-be-delivered sow; the camera monitors and obtains video data of a delivery room, continuously stores the video data to the video storage unit on one hand, and is directly connected with the server on the other hand; the server calls the backup video data and analyzes the monitoring data in real time; the working steps of thedetection system are as follows: monitoring the postures of the sows in real time, simultaneously detecting three areas of an approval area, a lactation area and a delivery area through a convolutional neural network area identification model, identifying the sows as lying postures when more than two areas are simultaneously detected, and outputting an identification result to a database. Comparedwith a method for recognizing the posture of the sow through a sensor technology, the computer vision technology avoids contact with the sow, stress response is reduced, and the method has the advantages of being low in cost and high in efficiency.

The invention discloses an efficient multistage anomaly flow detection method based on a TCP. A multistage anomaly detection mechanism is added in a traditional anomaly flow detection process. The method is used for anomaly detection for data flow sent by a client side in the network, the difference mean value method is used for carrying out difference stabilization processing on original flow produced by the client side, meanwhile, analysis and statistics are carried out on existing flow in the network, a self-adaptive threshold value interval is dynamically set, self-adaptive threshold value difference flow detection is carried out on the stabilized flow, and further anomaly detection is carried out on a data package which passes primary detection. The further anomaly detection is mainly used for analyzing the data package transmitted by a router, the key field is extracted, and whether the data package sent by the client side is abnormal or not is judged further by judging the key field. The efficient multistage anomaly flow detection method improves detection precision, and is easy and convenient to implement.

The invention discloses a method for identifying casting DR image loose defects based on an improved YOLOv3 network model. The method comprises the following steps: 1) performing defect labeling on aloose defect data set by utilizing a rectangular frame of an image labeling tool; and 2) establishing an improved YOLOv3 network model; 3) training the improved YOLOv3 network model by using the loosedefect data training set; 4) testing the trained improved YOLOv3 network model by using the loose defect data test set; 5) improving the improved YOLOv3 network model, and 6) obtaining a DR image ofthe casting to be detected, inputting the DR image into the improved YOLOv3 network model, and judging the defect grade and position coordinates of the casting. According to the invention, detection effects of small target objects by a target detectio network are improved.

The invention provides a transformer substation insulator infrared image detection method based on artificial intelligence. The transformer substation insulator infrared image detection method comprises the following steps: acquiring an infrared image of a transformer substation insulator through an infrared thermal imager; preprocessing the acquired image through an algorithm; performing target label processing on the obtained data set; dividing the data set into a training set and a test set; constructing an infrared image detection model of the improved feature fusion single-shot multi-boxdetector; carrying out training and parameter adjustment of the model by using a training set in the data set; performing target detection on the trained model by using a test set in the data set so as to prove the effectiveness of the model; through the above steps, automatic detection of the infrared image of the transformer substation insulator is achieved. By adjusting the parameters of the model and adding a feature enhancement module, the feature extraction capability of the model for the insulator is improved, and therefore it is ensured that safe and real-time detection is conducted onthe infrared image of the transformer substation insulator.

The invention provides a multi-QR code simultaneous extraction and detection algorithm. The algorithm includes the steps of first, processing an image by using a Canny algorithm; extracting an edge graphic of the image; searching in the edge graphic for contour graphics for positioning, and obtaining a nested relationship of the contours; after the contours are obtained, taking the three nested contours as a position detection graphic of a QR code; calculating the center distances of the contours as positioning points; and then, classifying the positioning points according to the relationship that the position detection graphics in the same QR code meet a isosceles right triangle, and finally extracting the QR codes. The algorithm is easy to implement, high in positioning accuracy, capable of meeting the requirements of real-time detection, high in accuracy, fast in identification, and capable of good use in a deployment and production environment.

The invention provides a fabric defect detection method based on a deep separable convolutional neural network, and the method comprises the steps: firstly marking a collected fabric defect image through a marking tool as a fabric image data set, and dividing the fabric image data set into a training set and a test set; secondly, constructing a depth separable convolution module, and constructinga DefectNet network by using the depth separable convolution module; inputting the training set into a DefectNet network for training, and adjusting parameters of the DefectNet network by using a training strategy of asynchronous gradient descent to obtain a DefectNet network model; and finally, inputting the fabric image in the test set into a DefectNet network model to obtain a target defect anda position coordinate in the image, and framing a defect target in the image. According to the method, deep separable convolution and multi-scale feature extraction are combined to build the convolutional neural network model, the detection precision is very high, the detection speed is greatly improved, and the requirement of real-time detection is met.

The invention relates to a bamboo cane surface defect detection method based on a triple loss network. The method comprises the following steps: S1, collecting bamboo cane surface defect data througha camera mounted on a bamboo cane sorting robot to form a bamboo cane surface defect data set; S2, performing triple loss network training on the images in the bamboo cane surface defect data set to obtain an anchor point prediction matrix, a size prediction matrix and a thermodynamic diagram prediction matrix; S3, calculating anchor point prediction matrix loss, size prediction matrix loss and thermodynamic diagram prediction matrix loss to obtain total loss of triple loss network updating; S4, continuously updating and optimizing to obtain an optimal convolution weight parameter and an optimal offset parameter based on the updated total loss of the triple loss network calculated in the step S3; and S5, obtaining an anchor point prediction matrix and a size prediction matrix by updating the triple loss network of the convolution weight parameter and the offset parameter of the test image in the bamboo cane surface defect data set, and then obtaining the category and the size of the detected target defect.

The invention belongs to the computer vision technical field and provides a target feature detection method and device. The method comprises the following steps that: a LBP (Local Binary Pattern) feature descriptors are extracted from a video frame; according to a preset simplification algorithm, a HOG (Histogram of Oriented Gradient) feature descriptors are extracted from the video frame; and a target can be identified through an SVM (Support Vector Machine) classifier according to the LBP feature descriptors and the HOG feature descriptors. According to the target feature detection method and device provided by the invention, the LBP feature descriptors and the HOG feature descriptors of the video frame are extracted according to the preset simplification algorithm so as to form HOG_LBP feature descriptors, and the HOG_LBP feature descriptors are inputted into the SVM classifier, so that a specific target in the video frame can be identified. The target feature detection method is suitable for FPGA (Field Programmable Gate Array) hardware circuits, and therefore, requirements for real-time detection can be satisfied, and computing speed of target feature detection can be improved.

The invention provides a spectrum sensing method based on polarizability, which is suitable for cognitive radio networks. The method comprises the following steps of: setting a system detection threshold, continuously sampling space electromagnetic signals, estimating a covariance matrix of the received signals, obtaining the polarizability of the received signals based on a characteristic value of the covariance matrix, thereby, judging existence of electromagnetic signals according to the polarizability difference between signals and noises. The method can realize application of polarization information of electromagnetic signals of authorized users, and effectively solves the problem of indeterminacy of noise in practice; the polarizability is obtained based on a sampling covariance matrix of the received signals, and detection can be performed without using the prior knowledge of signals, channels or noises; the method is suitable for small sample detection, and can reduce computational complexity, improve detection speed of the system to the electromagnetic signals and reduce detection time of the system to the electromagnetic signals. Thus, the method can meet the requirement on real-time detection, and is beneficial for efficient usage of limited radio spectrum resource.

The invention relates to a river surface garbage real-time detection method and device based on an unmanned ship and a storage medium. The method comprises the following steps: acquiring a river surface real-time video image; performing garbage target detection on the video image by adopting a pre-trained garbage recognition model to obtain the type of garbage and the position information of the garbage; and based on the type of the garbage and the position information of the garbage, determining the garbage distribution condition of the river surface so as to indicate the unmanned ship to salvage the garbage. The problems that current river surface garbage recognition is poor in real-time performance and the position and the type of garbage cannot be accurately recognized can be solved.

The invention discloses a binocular visual sense based intelligent obstacle avoidance algorithm. The algorithm comprises the following steps that S1) a route is planned according to a destination of arobot, left and right cameras of a binocular camera are calibrated, and plane equations of the ground in coordinates of the left and right cameras are calculated respectively; S2) the binocular camera collects an image in front of the robot, and the image is de-noised; and S3) distortion and polar lines in the two images collected by the left and right cameras of the binocular camera are corrected, distortion is eliminated, matching points are limited in one straight line, and finally, a parallax image is obtained by matching. The algorithm satisfies the requirement for detecting barriers correctly in high precision, the requirement of real-time detection can be met, the accuracy and instantaneity are both taken into consideration, and the robot achieves a good obstacle avoidance effect.

The invention discloses a belt longitudinal tearing detection method and device and a storage medium. The method comprises the steps of acquiring a plurality of tearing belt images and a plurality ofnormal belt images; preprocessing the acquired tearing belt images and the normal belt images to obtain clear tearing belt images and the normal belt images, taking the clear tearing belt images as positive samples, and taking the clear normal belt images as negative samples; extracting Haar features of each image in the positive sample and the negative sample, and constructing a plurality of weakclassifiers according to the Haar features; iteratively combining the plurality of weak classifiers to obtain a plurality of strong classifiers, and combining the plurality of strong classifiers to obtain a cascade classifier; and adopting the cascade classifier to identify the belt images collected in real time so as to detect whether belts are longitudinally torn or not. According to the invention, the problems of low belt tearing detection accuracy and poor illumination robustness in the prior art are solved.

The invention discloses a rope skipping number counting method based on human body posture estimation and a TPA attention mechanism. The rope skipping number counting method comprises the steps that 1, extracting human body key points in a rope skipping action video by means of an Openpose model; 2, obtaining an oscillogram of the distance between the key point and the reference line with respectto time; 3, constructing and training an SRNN model based on a TPA attention mechanism; and 4, comprehensively considering the judgment condition 1 and the judgment condition 2 to judge whether a ropeskipping action exists. According to the method, the number of skipping ropes of a detected object can be detected in real time through equipment, so that the detection precision and efficiency are effectively improved.

The invention relates to a spectrum detecting method which belongs to the technical field of communication. The method comprises the following steps: defining a system detection threshold value; continuously sampling spatial electromagnetic signals; overlapping and grouping the sampled data; confirming self-correlation values of all groups and merging the self-correlation values; establishing a covariance matrix; and judging if an electromagnetic spectrum signal exists. By using the method, a data signal received from a radio frequency front end is directly processed in the frequency-conversion sampling manner, the sampled data is grouped in preset mode, and the traditional serial data processing method is changed into the parallel data processing method, thereby increasing the running processing speed. By using the method provided by the invention, under the condition of maintaining the original detection property of system, the speed of the system for detecting the electromagnetic spectrum signal is increased, the time of the system for detecting the electromagnetic spectrum signal is reduced, the demand of real-time detection is satisfied, thereby being beneficial to efficiently using the finite radio spectrum resource by the system and realizing the spectrum resource sharing between radio users.

The invention relates to a method and a system for realizing intelligent detection of construction isolation fence state based on depth learning technology, the method includes: acquiring a real-timeimage as a monitoring image from a video monitoring system, establishing a depth neural network model for classifying the construction isolation fence states, monitoring and processing the construction isolation fence, classifying the monitored images and classifying and judging the states, and starting an alarm for the abnormal states found by the construction isolation fence. The method and thesystem for intelligently detecting the state of the construction isolation fence based on the depth learning technology can accurately distinguish the normal state from the abnormal state of the construction fence. When the depth neural network model is trained, the classification and detection of construction isolation fence only need non-iterative forward operation, which can meet the requirements of real-time detection. The intelligent detection method of the construction isolation fence state provided by the invention has the advantages of fast operation speed and high detection precision,so that the timeliness of the alarm can be ensured and the hidden danger of the construction can be strictly prevented.

The invention discloses a method of detecting a moving target under a static background, comprising the following steps: S1, using an optimized Canny edge detection algorithm to quickly and simply build a stable background edge model and a background update method, using a background edge difference method to perform differential operation on a current frame edge detection result and the built background edge to get a background edge detection difference result, and determining the edge of a moving target; S2, performing OR operation on the background edge detection difference result and a result of detection using a five-frame difference method to get a more complete contour of the moving target; and S3, performing assimilation filling on the holes inside the contour of the moving target by using a horizontal-vertical bidirectional template, and performing morphological processing and connectivity detection on the target to eliminate small isolated points and fill small gaps of the edge, in order to extract a complete and accurate moving target area.

The invention provides an electrical equipment infrared image real-time detection and diagnosis method based on lightweight deep learning. The electrical equipment infrared image real-time detection and diagnosis method comprises the following steps: S1, acquiring an infrared image of electrical equipment of a transformer substation through an infrared thermal imager; s2, preprocessing the obtained infrared image through an algorithm to form a data set for training; s3, performing target label processing on the obtained normal and fault data sets of the electrical equipment; s4, randomly distributing the processed data set into a training set and a test set; s5, constructing an infrared image real-time detection and diagnosis model of the improved lightweight single-shot multi-box detector; s6, performing parameter adjustment and training of the model by using the divided training set; and S7, carrying out automatic detection and diagnosis on the trained detection and diagnosis model by using the divided test set so as to prove the effectiveness of the detection and diagnosis model. Through the above steps, real-time detection and diagnosis of infrared images of various electrical devices (especially effective schemes can be deployed in limited environments such as embedded devices) are realized.

The invention relates to a lane line detection method and system for improving color spaces and search windows, and the method comprises the steps: obtaining a lane line region aerial view based on the perspective transformation of a lane line region image, and carrying out the re-fusion of the lane line region aerial view through employing different color components of different color spaces, and lane line searching and lane line fitting are carried out in combination with a window sliding mode, and final lane line detection is realized. According to the scheme, optimal channels in multiple color spaces are fused, and the problem that illumination, shadow and yellow lane lines are difficult to detect is effectively solved; repeated calculation steps can be ignored by applying a window search method, the calculation amount is effectively reduced, and the speed is increased; besides, based on the application of mutual transformation between perspective transformation and inverse perspective transformation, a large amount of calculation related to lane line detection is applied to an image with a small data size after perspective transformation, the calculation amount can be effectively reduced, the detection speed of a single-frame image is increased, and the requirement of real-time detection is met.