34results about How to "Overcome missed detection" patented technology

The invention discloses a detection method and system targets in a video. The method includes executing the following steps on each frame of image of the video until final detection results of the targets in all frames of images of the video are obtained, wherein the steps include: detecting targets in the current frame of image, and acquiring targets, of which confidence degrees are greater thana first threshold value, to use the same as an intermediate detection result of the current frame of image; and utilizing a final detection result of a previous frame of image to carry out fusion matching, which is based on corresponding of the same target in the previous video frame of image and the later video frame of image, on the intermediate detection result of the current frame of image, and obtaining a final detection result of the current frame of image. The method avoids missed detection of the targets of multiple video frames through fusion matching which is on previous frames and later frames of the video and is based on target detection results of the images.

The invention relates to an ultrasonic-acoustic emission detection method for a composite material. The method can detect internal defects of the composite material according to an acoustic emission signal excited by pulse ultrasonic waves in the composite material; an emission transducer and a receiving transducer adopt an asymmetric mode and two layouts such as a separation type and an integration type and are arranged on the same side or any side of a part of the composite material; the positions and the attitudes of the emission transducer and the receiving transducer are selected within 0-360 degrees, so that multi-channel ultrasonic-acoustic emission detection can be realized; defects can be judged and positioned according to a time domain feature of an acoustic emission signal in the detected composite material. According to the method, the composite material detection capability of ultrasonic-acoustic emission is greatly improved; the detection thickness is 50mm, and the detection sensitivity to the defects which have the diameters being phi3mm and are 0.13mm and 49mm deep is achieved; the ultrasonic-acoustic emission detection resolution, a surface detection blind area and the detection capacity are obviously improved.

The invention discloses a human body recognition tracking method, which comprises the following steps: the step 100, acquiring original video stream data, and converting the original video stream data into pictures to establish an initial data set; the step 200, performing enhancement processing and screening on the initial data set to obtain a training set, a verification set and a test set; the step 300, building a Centernet network structure composed of a backbone network, an up-mining path and a top convolution, wherein the top convolution adopts a deep separable convolution; the step 400, designing a BOX matching mechanism and a loss function to construct a complete Centernet network structure; the step 500, using the training set, the verification set and the test set to train, verify and test the complete Centernet network structure to obtain a Centernet network model; and the step 600, identifying and tracking the human body in the real-time video stream data by using the Centernet network model. According to the human body identification tracking method, the Centernet network structure is optimized, the detection speed is improved under the condition that the detection accuracy is not reduced, and the balance between the accuracy and the speed is optimized.

The invention discloses a strong convergence field recognition and speed correction method based on Doppler weather radar data, and the method comprises the following steps: designing a new coordinatesystem, converting radially distributed radar data into grid point data, and completing the extraction of a high-reflectivity area; designing a detection template, and obtaining a parameter vector for describing the convergence intensity and position of a convergence point on the basis that the convergence point is located quickly; extracting a slave convergence point to generate a convergence zone, and correcting the idea and algorithm of the convergence intensity through the direction of the convergence zone, thereby effectively solving a problem of the underestimating of the strong convergence field at a single pitch angle and even the detection leakage; determining a profile line according to the strongest convergence point at the single pitch angle, obtaining the better section of the strong convergence field, and obtaining the depth information of the strong convergence field. The method can achieve the automatic detection of the strong convergence field, solves the problems ofunderestimation and detection leakage to a certain degree through a speed correction method, achieves the automatic drawing of the section map of the strong convergence field, achieves the calculationof the feature parameter of the strong convergence field, and promotes the recognition and forecast of the convective weather disasters.

The invention discloses a rainfall detection device, a rain gauge and a rainfall determination method. The rainfall detection device comprises a capacitive rain sensing module which is used for detecting the rainfall condition and converting the detected rainfall change into a capacitance value; a piezoelectric rainfall sensing module which is used for detecting impact potential energy generated by raindrops and converting the impact potential energy into an analog signal; and a controller which is respectively connected with the capacitive rain sensing module and the piezoelectric rainfall sensing module, and is used for detecting rainfall according to the capacitance value transmitted by the capacitive rain sensing module and the impact potential energy transmitted by the piezoelectric rainfall sensing module. Through fusion of the two rainfall measurement modules, advantage complementation is realized, the rainfall detection accuracy and reliability are improved, and rainfall full-coverage detection is realized.

The invention provides a remote sensing image cloud detection method based on cascade color and texture feature attention. The method comprises the following implementation steps of: generating a training set; constructing a remote sensing image cloud detection network; training a remote sensing image cloud detection network; and detecting a cloud-containing remote sensing image. According to the method, a color feature extraction subnet and a texture feature extraction subnet are utilized, the problem that thin cloud features are difficult to extract due to the fact that shallow features extracted in the prior art lack the difference between cloud and ground features is solved, and the detection accuracy of non-boundary thin clouds is improved. According to the method, the color feature attention module and the texture feature attention module are utilized, the problems of large false detection and missing detection of boundaries and thin cloud areas in the prior art are solved, the thin cloud detection accuracy is improved, the missing detection rate and the false detection rate are reduced, and the overall accuracy of cloud detection is improved.

The invention relates to a smoothness-characteristic-quantity-based surface gray scale defect detection method for a rotary-cut wooden product. The method comprises the following steps: acquiring an image of the rotary-cut wooden product by using CCD (Charge Coupled Device) cameras placed at the middle upper positions of two LED (Light-Emitting Diode) strip light sources which are arranged symmetrically and adopt high-angle illuminating ways serving as illumination light sources, converting the acquired image into a digital signal through an image acquiring card, and transmitting the digital signal to a computer system; performing target area determination, smoothness characteristic quantity calculation and gray scale defect judgment on the input image, wherein the judgment threshold value of each gray scale defect is determined by using a self-adaptive maximum interclass variance method. By adopting the detection method, automatic online detection on the surface gray scale defects of the rotary-cut wooden product is realized, the production cost is reduced, the detection efficiency is increased, the detection rate of the gray scale defects is increased greatly, and the phenomena of detection missing and error detection are well overcome.

The present invention discloses a real-time automatic threshold calculating method for scanning and monitoring a radio signal, which relates to a calculating method in radio monitoring software. According to the real-time automatic threshold calculating method, according to real-time scanning on one frame of real-time scanning frequency domain monitoring data [Kn], a threshold value [T'n] can be quickly calculated through a real-time automatic threshold calculating method. According to the real-time automatic threshold calculating method, a counting automatic threshold calculating principle is adopted, the threshold value is based on certain time (or number of scanning); detection leakage may occur for the signals with short appearing time, and influence is caused for subsequent data processing (such as occupation rate counting). The real-time automatic threshold calculating method provides a signal determining basis for radio real-time scanning monitoring and overcomes the problems of: incapability of considering large signals and small signals by linear threshold, incapability of differentiating signals and noises, and detection leakage which may be caused by automatic threshold counting.

The present invention relates to the biotechnology field, and relates to entamoeba histolytica Gal/GalNAc polypeptide fragment and a preparing method and an application of the polypeptide fragment. The preparing method comprises the steps of: extracting total RNA from cultured entamoeba histolytica trophozoite, obtaining C terminal polypeptide fragment coding gene of Gal/GalNAc suppressible agglutinin of recombinant 150kDa through amplification, constructing expression vector, transferring to Escherichia coli for induction, realizing high expression of C terminal polypeptide fragment of the Gal/GalNAc suppressible agglutinin of the recombinant 150kDa in form of inclusion body, washing to obtain high-purity protein, and using glutathione reduction system for carrying out renaturation to obtain the C terminal polypeptide fragment of the Gal/GalNAc suppressible agglutinin of the recombinant 150kDa, wherein the polypeptide fragment possesses biological activity and solubility. The polypeptide fragment can be used for serology immunodiagnosis of amoebiasis, and overcomes detection miss and misdiagnosis of entamoeba histolytica cyst and trophozoite in stool examination in amoebiasis diagnosis, thereby obviously improving the specificity and sensitivity of the amoebiasis diagnosis.

The invention provides a ring object counting method based on Hough circle transformation. The method comprises the steps that inner ring parts of a target in an image to be counted is extracted; Hough circle transformation is used to screen the inner ring parts in the inner ring image; and finally, the number of screened inner rings is the final count of ring objects. According to the invention,the calculation of the number of ring objects is converted into the calculation of the number of inner rings of the image, which completely avoids a cumulative error caused by the indirect relationship that number is equal to total pixel points/single pixel point; the problems of difficult Hough circle transformation threshold setting and miss detection are overcome; fast counting is realized; thecounting efficiency and accuracy are improved; and high-accuracy fast counting of ring objects which have the characteristics of light weight, small volume and uneven density, such as an industrial sealed felt pad, is realized.

The invention discloses an ultrasonic-based and electromagnetic ultrasonic combined dual-mode nondestructive testing method and device. The method comprises the following steps: coarsely testing a metal member under the action of a non-static magnetic field by utilizing an ultrasonic detection method; and applying pulsed excitation current to the metal member under the action of a static magnetic field. Lorentz force is generated inside the metal member under the action of the static magnetic field, so that an acoustical signal is generated due to the Lorentz force, and ultrasonic signals generated at positions with defects or without defects of the metal member are different. The electric conductivity distribution of the metal member is reconstructed through an ultrasonic probe array attached to the metal member, and the defects of the metal member are detected according to the result generated by combining the electric conductivity distribution and the ultrasonic detection. According to a device by using the detection method, a control system (2) is connected with a pulsed excitation system (1), an ultrasonic probe array (4) and a magnet system (3); the ultrasonic probe array (4) is connected with a signal detection processing system (5); the signal detection processing system (5) is connected with a reconstruction system (6).

The invention discloses an intelligent substation relay protection hard pressing plate checking method based on a target detection model, and the method specifically comprises the steps: downloading a reference file corresponding to a work task, scanning a screen cabinet two-dimensional code, and building a relation with the reference file; after photographing is completed, calling of the YOLOv4-tiny model and analysis of the reference file are synchronously carried out, and a result checked by a worker is finally displayed; if missing detection and false detection do not occur, a worker can directly store the record to the local; and if the phenomenon of missing detection or false detection occurs, a worker needs to correct the corresponding hard pressing plate. According to the method, the protection screen cabinet does not need to be additionally transformed, the hard pressing plate picture shot on site is used for manufacturing the data set, the data enhancement algorithm is researched and used for expanding the data set, the problems of missing detection, error detection and the like in the prior art are further solved, and the method has higher universality and robustness.

The invention relates to a multi-specification text collaborative positioning and extracting method. The method comprises the following steps of 110, acquiring text image data; 120, performing image normalization operation; 130, separating image background information and foreground information needed to be collected; 140, performing global collaborative search, and extracting text block regions in a preset format; 150, performing local optimization search, and optimizing the positions of the text regions one by one within a small range; and 160, outputting text block positioning results, andproviding the text block positioning results for a subsequent individual character segmentation and recognition module. According to the method, the technologies of image processing, target detection,collaborative searching, local optimization and the like are utilized; the requirements of formatted data collection are met; the problems of missing detection and wrong detection after degradation of certain text blocks are solved, and unnecessary interference of text and noise information is avoided to the maximum extent; and the accuracy and precision of formatted text information collection are improved.

The invention discloses a pedestrian three-dimensional detection tracking method and system based on a top visual angle. The method comprises the following steps: performing stereo correction and stereo matching processing on left and right images acquired by a binocular camera; filling holes and converting to obtain a height map; modeling the height map to obtain a two-dimensional height foreground map; projecting and converting into a two-dimensional plane point cloud projection drawing under a top view angle, and detecting to obtain a three-dimensional coordinate point of the head of the pedestrian; mapping the three-dimensional coordinate points of the heads of the pedestrians back to the two-dimensional image coordinate system, and positioning a bounding box of each pedestrian; and performing inter-frame pedestrian matching tracking by combining the three-dimensional coordinate point of the pedestrian head and the positioned pedestrian bounding box. According to the invention, the problem of missing detection caused by shielding among pedestrians can be effectively overcome, and the recall rate of pedestrian detection is improved. The pedestrian three-dimensional detection tracking method and system based on the top view angle can be widely applied to the field of pedestrian detection.

The invention discloses a potential safety hazard detection method and device, electronic equipment and a readable storage medium, and is applied to the technical field of image processing, and the potential safety hazard detection method comprises the steps: obtaining a construction site image, and carrying out the potential safety hazard discrimination of the construction site image, so as to judge whether there is a potential safety hazard in the construction site image; and if potential safety hazards exist, carrying out potential safety hazard detection on the construction site image, and obtaining a potential safety hazard detection result, wherein the potential safety hazards comprise at least one of fire potential safety hazards, white potential safety hazards and collapse potential safety hazards. The technical problem of low detection accuracy of potential safety hazards in a construction site in the prior art is solved.

The invention belongs to the technical field of biomedicine and clinical application, relates to a polypeptide fragment specifically binding to the serum of Babesiosis patients, and an amino acid sequence thereof. According to the present invention, a series polypeptide fragments are obtained by analyzing the structure and the epitope of a Babesia merozoite surface protein BMASA, and the polypeptide fragment is prepared through polypeptide synthesis, and can be used for preparing the serological immunodiagnostic reagent for Babesiosis; and with the polypeptide fragment, the polypide missing diagnosis and the polypide misdiagnosis of the blood smear in the diagnosis of Babesiosis can be overcome, and the specificity and the sensitivity of the diagnosis of Babesiosis can be significantly improved.

The invention discloses a lineament extraction method of a remote sensing image, comprising the following steps of: (1) inputting a remote sensing image and carrying out the preprocessing on the image enhancement; (2) establishing a multi-resolution structural image sequence with a pyramid structure; (3) extracting lineaments in different levels for all levels of images with different resolutionsaccording to the consistency of gradient direction and the consistency of the regional gray; and (4) establishing characteristic database for the extracted lineaments and managing the extracted lineaments. In the method, by carrying out the pyramid classification on the remote sensing image and extracting different lineaments, the application range can be widened; by fusing the gradient directionof the lineaments with the gray linear characteristics, the precision for recognizing the linearity can be improved; the calculated quantity is reduced and the leaking detection and the false detection are overcome by using a fixed area while the lineaments are generated; and the lineament extraction method has favorable robustness and accuracy.