44results about How to "Object Detection Implementation" patented technology

The invention discloses a target space knowledge and two-stage prediction learning-based target detection method. By utilizing various data conversion methods, a sample number is increased and samplediversity is improved; two deep neural networks including SSD and newly designed RefineNet are trained; for a prediction target with a relatively high probability in a primary prediction result of theSSD, the accuracy of judgment is further improved through the RefineNet; and by establishing peculiar spatial structure constraint rules of the target, the wrong prediction is reduced, thereby obtaining a final detection result. Compared with a few existing methods, the method provided by the invention has the advantages that visual and spatial characteristics of a remote sensing target are considered at the same time; and end-to-end target candidate selection, feature extraction and classified locating are realized by utilizing the deep networks with excellent feature extraction capabilities, so that the detection rate of the remote sensing target is remarkably increased and the false alarm rate is reduced.

The invention belongs to the technical field of communication, and discloses a clutter inhibition method of an airborne external radiation source radar. The method comprises the following steps: obtaining signals received by the airborne external radiation source radar, wherein the signals comprises reference signals, first echo signals and second echo signals; respectively segmenting the reference signals, the first echo signals and the second echo signals to obtain segmentation reference signals and two paths of segmentation echo signals; according to the segmentation reference signals, respectively performing range-direction compression on the two paths of segmentation echo signals to obtain range-direction compression signals of the two paths of segmentation echo signals; respectively performing phase compensation on the range-direction compression signals of the two paths of segmentation echo signals to obtain phase compensation signals of the two paths of the segmentation echo signals; and performing channel registering on the phase compensation signals of the two paths of the segmentation echo signals, and performing subtraction on the two paths of the segmentation echo signals after the registering so as to inhibit clusters and reserve object information. The method provided by the invention solves the problem of neglect of incomplete cancellation of the clusters during migration of the reference signals in the prior art.

The invention discloses a vessel target real-time detection method based on a deep neural network. The method comprises the following steps: firstly, establishing a real-time deep neural network modelfor small target detection; constructing a small target training sample set according to a preset initial training sample set, and determining the optimal size range of the target; performing ROO training to obtain an initial deep neural network model; performing oHEM training; performing ship target detection on a preset initial training sample set by using the initial deep neural network model,adding difficult negative samples appearing in detection into the difficult negative sample set, and training the initial deep neural network model by using samples in the difficult negative sample set to obtain an optimized deep neural network model; and finally, establishing a remote sensing image pyramid model, and carrying out ship target detection layer by layer from the bottom layer of thepyramid by utilizing the optimized deep neural network. According to the method, the detection speed and precision can be greatly improved.

The invention discloses a frequency modulation broadcast outer radiation source radar object detection method based on waveform cognition. The method comprises steps of first recovering direct wave signals of different broadcasting stations in a reference channel; performing radiation source waveform cognition processing in the reference channel and selecting a radiation source; using a self-adaption compensation method to suppress the direct wave and the multipath clutter for signals received by a receiving array in a monitoring channel; and performing distance Doppler cross correlation calculation on compensation residual signals and direct wave signals of the selected radiation source in the reference channel, to achieve screening of the frequency modulation broadcast signal for objectdetection. Therefore, the technical problem that the bandwidth of the partial frequency modulation broadcast signal is not suitable for the object detection.

The invention relates to a high-resolution optical remote sensing image target detection method based on a rotation invariant HOG feature. The method comprises steps of: firstly, selecting a target image and a background image to obtain an initial training sample set, rotating the initial training samples according to given rotation transformation, and merging un-rotated training sample set with the rotated training sample set to obtain a total training sample set; training a rotation invariant HOG feature extraction module and a target classifier by learning a three-layer fully connected network, wherein a conventional HOG feature is the input of the first layer, the second layer is used for computing the rotation invariant HOG feature, and the third layer is a softmax classifier. The method solves a problem, by learning rotation invariant HOG feature, that it is difficult for the conventional HOG feature to process remote sensing image target rotation changes, can realize remote sensing image target detection, and achieves high detection precision.

The invention provides a hyperspectral image target detection method based on tension linear discrimination analysis dimension reduction. The objective of the invention is to solve problems that in the current hyperspectral image target detection method, characteristics of spatial constraint enhancement under the condition of high scores are not fully considered, information excavation cannot be performed on the whole three-dimensional information and the detection precision is quite low. The method comprises steps of 1, acquiring three-order target tension blocks, three-order background tension blocks and three-order to-be-detected test sample tension blocks; 2, allowing sub-space after the projection of the target tension blocks, the background tension blocks and the to-be-detected test sample tension blocks to have the biggest separability; 3, projecting the target tension blocks, the background tension blocks and the to-be-detected test sample tension blocks to a tension sub-space with the biggest separability; 4, calculating the total distance from each to-be-detected test sample to the background and the target; and 5, setting a threshold value and if the gray scale value is larger than the threshold value, determining the pixel of the central point as the target, or else, determining pixel of the central point as the background. According to the invention, the method is applicable to image processing field.

The invention discloses a three-dimensional detection method for a target in ground wave radar sea clutter, and the method comprises the steps: constructing a ground wave radar channel-distance-Doppler echo spectrum, extracting a sea clutter region, and constructing a multi-stage spatial domain filter through an oblique projection technology; feeding the ground wave radar channel-distance-Dopplerecho spectrum to each spatial domain filter, taking each direction covered by the ground wave radar by each filter as a potential target direction, and filtering in sequence; and performing synthesisoperation of taking a minimum value on the spectrum data subjected to spatial filtering to obtain a synthesis spectrum, performing three-dimensional detection on a distance-Doppler-azimuth spectrum ofthe synthesis spectrum to suppress non-uniform sea clutters, searching for a peak value, and finally outputting azimuth, distance and Doppler parameters of a ship target. According to the invention,azimuth dimension detection is expanded on the basis of traditional distance-Doppler dimension target two-dimensional detection, azimuth-distance-Doppler dimension target detection is further formed,and the target detection capability in sea clutter is improved by combining echo differences of sea clutter and a target in the three combined dimensions.

The invention relates to a small target detection method based on multi-frame differential image accumulation. The method comprises the steps of S1, obtaining and keeping a standard background image ; S2, acquiring multiple frames of images to be detected; S3, respectively carrying out differential operation on each frame of image to be detected obtained in the S2 and the standard background image to obtain a differential image; S4, accumulating the differential images obtained in the S3; and S5, judging whether a to-be-detected target appears or not according to a differential accumulation result; if the to-be-detected target appears, ending the detection; and otherwise, accumulating the multiple frames of to-be-detected image sequences obtained in the step S2 to obtain an average value, taking the average value as a new standard background image and returning to the step S2. According to the method, influence of random noise on differential calculation can be effectively avoided, the difference between the target and the background is enhanced, the separation of the target and the background is further realized, and the target detection can be more accurately realized. Compared with the existing method, the method provided by the invention has higher detection sensitivity and accuracy in an application scene in which the target hovers and is close to the background gray scale.

The invention discloses a substation scene personnel target detection method, which comprises: inputting a to-be-detected image into a pre-trained brightness detection network, and obtaining brightness information of the to-be-detected image; determining a difference threshold according to the brightness information, obtaining a background-removed difference image by using an image difference method based on a fixed background reference, and determining personnel target scale information in the to-be-detected image; and inputting the to-be-detected image, the differential image and the personnel target scale information into a pre-trained personnel target detection network to obtain a position frame and a type of the personnel target. Meanwhile, the invention discloses a corresponding system. According to the invention, the background in the image is removed through the image difference method, and the personnel target detection network is adopted to effectively realize the personnel target detection of the substation scene with the complex background.

The invention discloses an SAR image target detection method based on multi-source feature migration and false alarm rejection, and the method comprises the steps: obtaining a to-be-detected image, and carrying out the detection of the to-be-detected image according to a preset false alarm rate and a two-parameter constant false alarm rate algorithm, and obtaining a plurality of target pixel point sets; the to-be-detected image is a synthetic aperture radar (SAR) image; inputting each target pixel point set into a region sensing model to enable the region sensing model to detect the target pixel point sets; rejecting or retaining a target pixel point set according to a detection result, and taking the retained target pixel point set as a target detection result; wherein the region sensing model is a pre-trained neural network model. According to the method, preliminary target detection is performed on the to-be-detected image through the two-parameter constant false alarm rate algorithm, and false alarm elimination is performed by using the region sensing model, so that the calculation cost can be reduced, the problem of relatively high false alarm rate of a detection result is solved, and target detection of the pixel level of the SAR image is also realized.

The invention discloses a high-resolution remote sensing scene target detection method based on a size balance FCOS. The method comprises the steps: carrying out the dynamic adjustment of a centrality coefficient according to the regression information of each target through the centrality and frame regression stage of a size balance coefficient in a target detection module of the FCOS, distributing a reasonable weight for the frame regression process of each positive sample, using a high-resolution remote sensing target to detect a remote sensing data set for model training, and using the model for remote sensing ground features for recognition. According to the method, the defects of targets of different sizes in an FCOS centrality evaluation system are fully considered, loss weight reinforcement is carried out on samples, distributed at the edge, of positive samples in a small target anchor frame, redundancy loss contribution in a large target is inhibited, and target size balance is achieved; and the size balance FCOS improves the precision of target detection in a high-resolution remote sensing scene under the condition of not introducing extra overhead to a model reasoning stage.

The invention discloses a target detection method, electronic equipment and a computer readable storage medium. The method comprises the following steps: starting from an initial region of an image to be processed, a convolution kernel slides in the image to be processed at a preset step length to obtain a plurality of regions to be processed, the image to be processed is composed of multiple types of slice pixels, and the distribution conditions of the multiple types of slice pixels in each region to be processed are the same; taking one type of slice pixels as to-be-processed slice pixels, and performing matrix inner product multiplication on the convolution kernel and each to-be-processed region to obtain an inner product result of each to-be-processed region; obtaining a slice image by using the inner product result of all the to-be-processed areas; modifying the category of the to-be-processed slice pixels, and repeatedly executing the steps until each category of slice pixels are traversed to obtain a plurality of slice images; and obtaining a target detection result of the to-be-processed image based on the plurality of slice images. In this way, the precision of target detection can be improved.