36results about How to "Achieve precise extraction" patented technology

The invention discloses a method for extracting a target object from an interactive video, comprising the following steps of: obtaining the frame difference image of three adjacent frames of images by a frame difference method as a rough estimate of the target object; then combining with the coordinate position of the rectangular region of the front frame of image to determine the rectangular region containing the target object in the present frame of image; after that, structuring the energy function of each frame of image in an original video, and realizing the accurate extraction of the target object; adequately utilizing the coordinate position of the rectangular region of the front frame of image during the process of determining the rectangular region containing the target object inthe frame of image, thus, the target object region of the present frame can be accurately determined even if aiming at the conditions of target object suddenly turning to static or background shakingand the like; and finally establishing a fraction model for each frame of image after the target object region is determined, wherein the target object can be accurately extracted by a graph cut method, thereby greatly improving the extraction accuracy.

The invention discloses a water body remote sensing identification method based on U-net. The method comprises the following steps: A, carrying out data collection and preprocessing; b, carrying out preliminary extraction of a water body; c, processing a water body extraction result; d, making training and verification samples; e, training a U-net neural network model; and f, extracting water body. Two water body preliminary extraction algorithms are set through spectral analysis of different ground objects in each wave band of an image, all water bodies and part of other ground objects can beextracted, and a public truth value area, namely a preliminary water body result, is reserved. Classification post-processing, format conversion, modification, training and verification sample makingare carried out on an extraction result, a model is obtained through training of a U-net neural network model, and a water body is predicted. The method is suitable for regions with complex surface structures, is simple to operate and high in universality, can better eliminate interference of shadows, buildings and farmlands in the field steeping period, has a good extraction effect on small lakes and small ponds, and realizes accurate extraction of water bodies.

The invention provides a building instance extraction method and device for a remote sensing image, a storage medium and an apparatus, and the method comprises the steps of extracting a feature map ofthe remote sensing image data, and estimating a building instance existing in the remote sensing image data and an initial detection frame of each building instance according to the feature map; determining a candidate boundary line set corresponding to the building instances according to the feature information of the feature points in the initial detection box of each building instance; selecting a target boundary line corresponding to the building instance from the candidate boundary line set according to the similarity between the prediction position area of the building instance in the initial detection frame of each building instance and the effective area of each candidate boundary line in the candidate boundary line set of the corresponding building instance; adjusting respectiveinitial detection frames according to the target boundary lines of the building instances; and extracting the corresponding building instances from the obtained target detection frames of the buildinginstances, and realizing the accurate extraction of the building instances in the remote sensing image.

The invention, which is suitable for the field of the image processing technology, provides an image segmentation method and apparatus. The provided image segmentation method comprises the following steps that: a sliding operation is detected and a closed curve encircling a target image is generated at an original image according to the sliding operation; a binaryzation image template is generated based on the closed curve, wherein the size of the binaryzation image template is equal to that of the original image; according to the generated binaryzation image template, detection is carried out and the contour of the target image is obtained; a mask is generated according to the contour of the target image; a transparency value of an unknown zone in the mask is calculated; a foreground component and a background component of the original image are determined by utilizing the transparency value and the original image; and on the basis of the determined foreground component and background component of the original image, a foreground image is obtained by segmentation from the original image. According to the invention, the operation feature and the processing performance of the mobile terminal are fully combined; and image segmentation can be realized effectively at the mobile terminal.

The invention belongs to the technical field of computer vision measurement, and relates to a high-precision restoration method of fuzzy image based on vision prior information and system reference quantity. According to the exposure time of binocular vision measurement system, The accurate prior information of frame rate is used to obtain the changing direction of the same center position in image coordinate system and the moving distance in exposure time, that is, the blur direction and the blur scale, so as to obtain the accurate blur kernel of each image by using the relationship between two adjacent frames in the image sequence. Then the non-blind image restoration algorithm is used to deconvolution and get the clear restoration image, which realizes the high-speed image deblurring and restoration in the machine tool dynamic detection, and improves the measurement accuracy of the vision measurement system. The method can effectively improve the motion blurring problem of serial images, and the deblurring effect is obvious, which meets the requirements of precise extraction of machine tool positioning information in high-speed motion state.

The invention discloses a near-shore aquaculture area remote sensing image extraction method based on multi-feature and spectrum fusion, and the method comprises the steps: firstly preprocessing an inputted original remote sensing image, extracting a feature spectrum through a feature index method, and forming a ground feature feature set; secondly, using an improved constraint energy minimizationalgorithm for enhancing a target ground object in a culture area and weakening background spectral information; then using an Otsu method for calculating a threshold value, and preliminarily extracting the target ground object in combination with a single-waveband threshold value; and finally, carrying out customized elimination on the preliminarily extracted result by applying a gray-level co-occurrence texture matrix or an object-oriented method according to the ground object characteristics, and outputting a final result of culture area extraction. Compared with a traditional target detection method, the method has the advantages that even in a breeding area with the obvious foreign matter same spectrum phenomenon, background ground object interference can still be effectively overcome, a high-precision raft breeding area extraction result can be obtained, and the high-precision extraction requirement of the raft breeding area can be well met.

The invention discloses a moving target detection method with the combination of region extraction and an improved texture feature. The method comprises the steps of (1) collecting successive images in a monitoring video as a sampling frame, (2) carrying out background modeling and recovery on each pixel in the sampling frame by using its sampling information, (3) segmenting the image, extractinga foreground region by using statistical features of image blocks, judging the illumination change in the foreground region, and determining whether the second judgment of the foreground region is needed, and (4) carrying out the accurate extraction of foreground pixel points in the foreground region. By performing fast foreground region extraction, the amount of calculation for subsequent accurate judgment is greatly reduced, while the region extraction is carried out, two main interferences including a spatial displacement interference (leaf sway, etc.) and a brightness variation interference (lighting change, etc.) are eliminated, and a moving target in the image sequence is accurately and efficiently extracted.

The invention relates to an inverter IGBT tiny fault feature extraction method based on multi-modal data. The method comprises the following steps: 1), building an equivalent second-order system of a transient mode of a two-level traction inverter, and carrying out the parameter identification of an expression of the equivalent second-order system; 2) carrying out modal division on the output voltage data of the inverter through an Elman neural network; 3) extracting corresponding tiny fault characteristic parameters in each transient mode, wherein the tiny fault characteristic parameters comprise a voltage overshoot sigma, peak time tp and a voltage slope absolute value delta; 4) extracting a voltage steady-state value Vstab in each steady-state mode; and 5) carrying out fault detection according to the extracted characteristic parameters of the IGBT tiny fault. Compared with the prior art, the method has the advantages of easy realization, cost saving, noise interference reduction, modal classification accuracy improvement, accurate extraction, fault detection and the like.

The invention relates to a hazardous chemical solution extraction system based on a mobile robot. The hazardous chemical solution extraction system comprises a total console system, a communication network and the mobile robot. The mobile robot comprises a moving trolley system, a multi-joint mechanical arm system, a clamp system, a vision ultrasonic testing system and a hazardous chemical solution storing system. By the adoption of the system devices, hazardous chemical solutions can be extracted accurately, the surplus hazardous chemical solutions can be treated automatically, the intermediate link is omitted, and the time cost, labor cost and management cost are reduced.

The invention discloses a hybrid DC transmission line traveling wave protection method and system based on dual tree complex wavelet transform, and the method comprises the following steps: calculating voltage difference modulus and voltage common modulus according to DC voltage of positive and negative electrodes collected; The dual-tree complex wavelet transform is used to calculate the voltagedifference modulus, and the protection is judged whether to start or not according to the high sensitivity start-up. When the protection is started, the fault type is judged and the corresponding protection action is carried out according to the dual tree complex wavelet transform of voltage common modulus and voltage difference modulus and the corresponding fault extremum by using the fault judgment basis of the hybrid DC transmission line. The invention can quickly distinguish the faults of the hybrid HVDC transmission line, can reliably carry out fault pole selection, greatly reduces the translation sensitivity in the common real wavelet transform, can remarkably improve the accuracy of fault analysis in the long-distance transmission line, and realizes the fast identification of the faults of the hybrid DC transmission line.

The invention provides a hemoglobin concentration prediction regression model training and application method, and system, and the method comprises the steps of obtaining a multi-channel PPG signal, and carrying out the signal processing of the multi-channel PPG signal; performing manual and automatic feature information extraction on the acquired multi-channel PPG signals, and fusing the manual feature information and the automatic feature information through a feature information fusion mechanism to realize accurate extraction of the feature information; establishing a hemoglobin concentration prediction regression model integrated based on a multi-machine learning algorithm for noninvasive detection, and obtaining a more stable detection result by utilizing efficient integration of a plurality of regression models and multiple regression predictions. According to the method, under the condition that the model training speed is guaranteed, the detection precision and generalization ability of the model can be improved, the method has the advantages of a non-invasive detection method and an invasive detection method, and the technical problem that the hemoglobin concentration of the human body cannot be continuously detected in real time in the prior art is solved.

The embodiment of the invention discloses an extraction method and device for OLED drawing, firstly, vertical projection processing is carried out on the collected pixel distribution map, and the pixel grid is generated, performing parity column separation processing on the pixel grids to obtain the odd column grid and the even column grid, performing transverse projection processing on the odd column grids and the even column grids respectively to obtain the odd column transverse grid and the even column transverse grid, fusing the odd column transverse grids and the even column transverse grids, generating the pixel effective area, and determining the pixel center of the pixel effective area, region division processing is carried out on the pixel effective area according to the pixel center, and obtaining the pixel map; by performing parity column separation processing on the pixel grid, the interval between row pixels is increased, region division processing is carried out on the pixel active region, so that all energy of the pixel can be covered by the result of the region division processing, and accurate extraction of the OLED image is realized.

The invention discloses a modulation depth extraction and compensation method in a PGC phase demodulation method. The method comprises the steps: multiplying filtered, amplified and analog-digital sampled interference signals by reference signals of a first-order harmonic, a second-order harmonic and a third-order harmonic respectively, performing low-pass filtering to obtain three harmonic amplitude signals, performing differential operation on the three harmonic amplitude signals to obtain three harmonic differential signals, and solving modulation depth by using the harmonic amplitude signals and the harmonic differential signals; in combination with a Bessel function recursion formula, constructing a new harmonic amplitude signal which is not influenced by the modulation depth throughthe harmonic amplitude signal and the solved modulation depth value, so that the influence of the modulation depth is eliminated; and finally, accurately solving the phase to be measured through arc tangent operation. According to the method, the problem that nonlinear errors are difficult to compensate in real time due to modulation depth fluctuation in the PGC phase demodulation technology is solved, the phase measurement precision is improved, and the method can be widely applied to the technical field of interferometric optical fiber sensors and sinusoidal phase modulation interference.

The invention belongs to the technical field of functional materials, and particularly relates to a rare-earth phosphonic acid ultrathin two-dimensional material as well as a preparation method and application thereof. The molecular formula of the rare-earth phosphonic acid two-dimensional material disclosed by the invention is [M(R-PO3)(R-PO3H)H2O], wherein M is any one or a combination of two ofrare earth Eu and Tb trivalent cations, R-PO3 is a completely deprotonated form of an organic phosphonic acid group of R-PO3H2, R-PO3H is a partially deprotonated form of the organic phosphonic acidgroup of the R-PO3H2, and R is m-carboxyphenyl or m-isothiocyano phenyl; and the two-dimensional material is obtained by carrying out one-step hydrothermal reaction on a rare-earth salt and the organic phosphonic acid. The rare-earth phosphonic acid ultrathin two-dimensional material aiming at fingerprint characteristic chemical substances can ensure accurate extraction of a fingerprint and the clear line of the fingerprint by specific reaction between the material and the mark of the fingerprint. The fingerprint extraction material provided by the invention is simple in preparation method andeasily available in raw materials, the fingerprint extraction operation is simple, the fingerprint color development is clear, and the fingerprint extraction material has an application prospect in criminal investigation.

The embodiment of the invention provides an image extraction method and device, relating to the image field and used for accurately extracting target images. The method comprises obtaining a first image, receiving the infrared light reflected by a reference body and obtaining a second image in dependence on the reflected infrared light; wherein the second image corresponds with the first image; the first image and the second image are sent to an image processing device, so that the image processing device maps the second image with the first image and extracts target images from the first image in dependence on the second image. The image extraction method and device are used for extracting images.

The invention discloses a full-scale fiber bragg grating piezoelectric tuning multi-gas sensing system and method. An optical fiber coupling lens group is connected with a branch formed by alternately connecting an optical fiber circulator and a fiber bragg grating in series; each fiber bragg grating is adhered to one piezoelectric ceramic piece, and all the piezoelectric ceramic pieces are electrically connected with the piezoelectric ceramic control circuit; the optical fiber circulator is connected with the multi-path optical switch, and the single-path output of the optical switch is connected with the optical fiber collimating lens group; the optical fiber collimating mirror group is arranged corresponding to the gas absorption tank, a right side cavity mirror of the gas absorption tank is connected with the stepping motor through a transmission shaft; the pre-amplification circuit is connected with a photoelectric detector and is connected with the transmission shaft through a connecting frame; the pre-amplification circuit is connected with the microcontroller through the A/D sampling circuit. According to the invention, a fiber bragg grating piezoelectric tuning method is adopted, a low-cost broadband continuous spectrum light source can be adopted, the wave band range is wide, and various gases can be detected at the same time.