40results about How to "Improve target detection rate" patented technology

The invention relates to a target candidate region extraction method based on an image background mask, comprising the following steps: (1) constructing an image background mask data set; (2) buildinga GAN model, and adding a background mask for an image by adopting a GAN training way; (3) defining a loss function, namely enabling a generated picture and a trained target picture to be similar tothe utmost extent when a detail part of high-frequency structural information in the image is processed, and defining the loss function to be a combination of a GAN target function and one-norm distance loss of a composite image; and (4) carrying out model training.

The invention discloses a method for identifying and detecting abnormal behaviors in a transformer substation based on artificial intelligence in a complex scene. The method comprises the following steps: processing a monitoring video to obtain a static graph; detecting a human body region by using a target detection algorithm FPN network based on deep learning; preprocessing the to-be-identifiedimage to generate a binary image; taking the binarized image as the input of a CPN network to detect the key points of the human skeleton; fusing the human skeleton key point image with the RGB single-frame static image, inputting the fused image into an LSTM network for classification and identification, and judging whether the behavior is an abnormal behavior or not. According to the invention,automatic detection and identification tasks of abnormal behaviors of the working area of the transformer substation in a complex scene are realized, and the method has good accuracy, stability and real-time performance, and can meet the actual application requirements of the transformer substation.

The invention relates to a method and apparatus for detecting infrared dim small targets under a complex background, belonging to the technical field of infrared target detection. At first, the invention adopts a bilateral filtering algorithm to preprocess the single-frame infrared images in an image sequence successively, so as to reduce noise interference in the image and enhance the signal-to-noise ratio at the target; then, a single-frame infrared target detection algorithm based on block extremum is used to detect the potential candidate target sets in each preprocessed image frame by frame; based on the spatio-temporal continuity of small targets in multi-frame images, a pipeline filtering algorithm is used to identify the possible real targets in the candidate target set; finally the false alarm removal strategy is used to eliminate the false target and finally confirm the true target. Through the process, the invention overcomes the shortcomings of weakening target intensity, blurring edge, enlarging contour and the like brought by the prior infrared image processing technology, and can better retain and enhance the dim small target information, improve the target detectionrate, and reduce the missed detection rate and false alarm rate of the target.

The invention discloses a detecting and tracking method, device and equipment of a target object in a video. The method comprises a step of inputting a continuous video frame into a convolutional neural network obtained by training in advance, wherein the convolutional neural network at least comprises a set number of shared convolutional layers and region proposal network layers, a step of extracting features of the continuous video frames by using the shared convolutional layers to obtain feature mapping maps corresponding to different video frames, a step of determines a target area associated with the target object according to the feature mapping maps by using the region proposal network layers, and a step of detecting the position and running trajectory of the target object in the continuous video frames based on the target area. According to the method, the detection and tracking are unified by using a convolutional neural network model, the amount of calculation is reduced, theproblem of detecting multiple poses and multiple angles of view of the target can be solved, the target detection rate is improved, and the false detection rate is reduced.

The invention discloses a node duty ratio self-adaptive setting method for efficient target monitoring. The method comprises the following steps: according to the distance condition between a node and a base station, a far base station adopting a large duty ratio, and a near base station adopting a small duty ratio. The residual energy at the far base station region is sufficiently utilized, and the duty ratio of the node in the far base station region is improved. The omission ratio and the detection delay in the target monitoring can be reduced when the perception duty ratio of the node is improved; and the transmission delay in the target monitoring can be reduced when the communication duty ration of the node is improved, thereby effectively improving the target monitoring quality, and improving the energy utilization rate of a sensor network.

The invention belongs to the technical field of radars, and discloses an SAR image ship target detection method based on superpixel statistical diversity. The method comprises the steps of obtaining an SAR image to be detected, and performing super-pixel segmentation on the SAR image to obtain W super-pixels; secondly, based on a gamma distribution hypothesis, calculating a shape parameter and aninverse scale parameter of gamma distribution corresponding to each superpixel; calculating a global contrast value and a local contrast value of each superpixel, further calculating to obtain a TCR enhancement value corresponding to the superpixel, and taking the TCR enhancement value as an intensity value of each pixel point in the superpixel so as to obtain a TCR enhanced image; and finally, calculating a detection threshold T, and binarizing the TCR image to obtain a binary image corresponding to the TCR enhanced image. The TCR can be better improved on the basis of superpixel statistics diversity, so that a relatively high target detection rate can be realized at a relatively low false alarm rate, and the target detection performance is improved.

The invention provides a wide and narrow beam cooperative satellite-borne AIS message real-time receiving and processing system, which comprises: a wide and narrow beam cooperative module, wherein a plurality of VHF frequency-band antennas form an antenna array, multiple wave beams form a network to enabling the antenna array to form a plurality of wide and narrow beams to receive AIS signals; anda multi-channel parallel processing module which adopts a multi-channel AIS receiver, wherein each channel processes the AIS signal received by a single wave beam, filtering, amplification and sampling of the signal, demodulation of a message and extraction of important information of a ship are completed, and meanwhile, the multi-channel parallel processing module has a signal time slot conflictresolution capability. By means of simultaneous receiving of wide and narrow beams, multichannel parallel processing, on-board extraction of important information, synchronization of message second information and satellite time and the like, the system has the advantages of comprehensively improving the ship target detection rate, extracting ship simplification information, obtaining accurate timestamps of messages and the like.

The present invention discloses a generalized eigen-decomposition-based full polarimetric high resolution range profile target detection method. The method comprises the following steps of (1) obtaining the training target echo and the training clutter as the training data according to the echo of a known full polarimetric radar; calculating a covariance matrix C (O) of the coherent vectors of the training target echo and a covariance matrix C (C) of the coherent vectors of the training clutter; calculating a projection matrix P; (2) obtaining the test full polarimetric high resolution range profile of the full polarimetric radar as the test data; dividing the test data into L range cells, and extracting a coherent vector of each range cell of the test data; carrying out the premultiplication on the coherent vector of each range cell of the test data and the projection matrix P to obtain a reconstructing coherent vector of each range cell of the test data, and calculating the two-norm of the reconstructing coherent vector of each range cell; setting a detection threshold eta, if the two-norm of the reconstructing coherent vector k 'D(1) of the first range cell is not less than eta, determining the test data as a target, otherwise, determining as the clutter.

The invention discloses a target detection method and device, computer equipment and a computer readable storage medium and belongs to the technical field of computer vision. The method comprises the following steps that a plurality of detection frames in a detection image in a video are determined according to a target detection model; according to a target frame in each image in first n images adjacent to the detection image in the video, a plurality of prediction frames in the detection image are determined, the target frames are used for indicating an area where a target exists, and n is an integer greater than or equal to 2; and a target frame is selected from the plurality of detection frames in the detection image according to a plurality of prediction frames in the detection image. The method is advantaged in that detection frame error filtering caused by target shielding, target crowding and the like can be avoided, and the target detection rate is improved.

The invention relates to a radar echo interference removing method. Echo signal obtained when no target passes through are taken as modulation leakage signals, the pre-detected modulation leakage signals are subtracted from detected target echo signals to obtain target echo signals. Even targets with low spectrum peaks can be detected, and accordingly the target detection rate is increased.

The invention relates to a visible light-based open fire smoke rapid identification method and system. The method comprises the following steps: carrying out graying processing on an obtained fire image to obtain a gray scale fire image; calculating texture feature information of the grayscale fire image by adopting a spatial grayscale layer co-occurrence matrix method; carrying out smoothing processing and gray stretching on the gray fire image; extracting a suspected flame shape in the image subjected to smoothing processing and gray stretching processing by adopting an edge extraction operator; calculating characteristic parameters of the extracted suspected flame shape; and judging whether the suspected flame shape is flame or not according to the obtained characteristic parameters. According to the invention, the detection speed and accuracy can be improved.

The invention provides a target classifier self-adaptive updating method and a target classifier self-adaptive updating device. The target classifier self-adaptive updating method comprises the steps of: carrying out target detection on an image to be detected by adopting a target classifier; recording position coordinates of component models when synthetic scores are greater than a score threshold value during the detection process; acquiring position coordinate standard deviations of the component models through establishing Gaussian models of the position coordinates of the component models; and updating anchor point coordinates of the corresponding component models according to the position coordinate standard deviations of the component models, thereby achieving self-adaptive updating of the target classifier. The method and the device can achieve self-adaptive updating of the target classifier in different scenes, and increase a target detection rate.

The invention provides a double-deletion threshold-based target detection method. According to the method, firstly, on the basis that a maximum rejection threshold value and a minimum rejection threshold value are set, the subordinative function value of a sample reference unit is introduced. Secondly, the subordinative function value is compared with the maximum value and the minimum value, and then a corresponding binary weighted value of the sample reference unit is obtained. Thirdly, an average value is taken to obtain a background noise power estimation value. Fourthly, the background noise power estimation value is multiplied by a scaling coefficient to obtain an optimal power detection threshold value. Finally, the power of a test unit is compared with the power detection threshold value. If the power of the test unit is larger than or equal to the power detection threshold value, a target is judged to exist. Otherwise, no target is judged to exist. The method solves the technical problems in the prior art that an existing detection method is small in detection rate and high in leakage detection rate and false alarm rate, and the determination of maximum and minimum rejection threshold values must be dependent on the priori knowledge. Meanwhile, the method is high in rejection performance, high in target detection rate, and small in leakage detection rate.

The invention discloses an improved pedestrian target detection algorithm in a Faster R-CNN tunnel environment, and the algorithm comprises the following steps: building a pedestrian target data set in a highway tunnel environment, and randomly dividing the pedestrian target data set into a training set and a test set; optimizing Anchor in the Faster R-CNN network by adopting an unsupervised learning algorithm on the basis of the training set obtained in the above steps to obtain anchor setting; establishing a cavity convolution pyramid structure; designing an attention mechanism for processing the feature information and enhancing the expression ability of the features; and establishing a pedestrian detection framework in an expressway tunnel environment. According to the method, the pedestrian target feature extraction capability under the conditions of dark images, small target relative scale, vehicle lamp influence and the like is improved, and the pedestrian target detection ratein a tunnel environment is improved.

The invention relates to a false target detection optimization method based on BSD radar left-right communication, and the method comprises the steps of: acquiring target lists detected by radars at two sides of a vehicle, and enabling the target lists to communicate with each other; screening and identifying the detection targets on a first side of the vehicle according to the target lists detected by the radars on the two sides of the vehicle to obtain a reference real target and an uncertain target; matching the reference real target with the uncertain target, if the reference real target is matched with the uncertain target, judging that the target detected by a second side radar of the vehicle is a real target, otherwise, judging that the target is a false target; and according to a matching result, rejecting false targets, and updating target list information detected by radars at two sides of the vehicle. According to the method, through mutual communication of the radars on the left side and the right side of the vehicle, the authenticity of targets detected by the radars on the left side and the right side of the vehicle is verified, false targets are eliminated, identification of the false targets is reduced, the target detection rate of the BSD radars is improved, and the method is rapid and efficient, can greatly reduce detection of the false targets and improves the detection rate of real targets.

The invention relates to the technical field of large hydropower engineering low-altitude defense systems, in particular to a bit synchronization algorithm for maximum likelihood estimation, which comprises the following steps: S1, defining a system defense area; s2, setting a distributed and multi-node defense system; s3, formulating alarm levels of multiple stages; s4, setting a dual-guarantee working mode; s5, information is comprehensively obtained through the system, and comprehensive evaluation is carried out; and S6, the equipment and the system perform self-inspection, and the system defense area is divided into two parts, namely a core area and a detection early warning area: a, the equipment can form an invisible dome-type protection cover in the core area, and the hierarchical management of the whole area can be completed through the distributed and multi-node defense system; according to the invention, the working efficiency of the system can be effectively improved, the operation process of the system is simplified, the reliability of the system is improved, the detection accuracy and precision of the target can be improved, and the purposes of accurate identification and accurate strike of the low-altitude target are achieved.

The invention provides an infrared target detection method and a computer storage medium. The infrared target detection method specifically comprises the following steps: inputting continuous video frame data; correcting lens shake and background change of continuous video frame data through an image registration method based on a foreground mask algorithm; associating adjacent frame targets in continuous video frame data by using time and space constraints; in combination with a KCF algorithm, further tracking the associated target to obtain a tracking result; and finally determining whether the target is a moving target according to the association frequency and the tracking result of the target. According to the infrared target detection method provided by the invention, shaking and background change are corrected through feature point matching based on KLT, adjacent frame targets are bound and associated by using features such as time and space; and then the multi-dimensional target association is carried out by combining a kernel correlation filtering tracking method based on KCF and a moving target is detected in a combined manner. Therefore, infrared weak and small target detection rate under airborne forward-looking and downward-looking conditions is improved, and the false alarm rate is reduced.