255 results about "Pixel classification" patented technology

The invention discloses a remote sensing image semantic segmentation method based on a double-branch multi-scale fusion network, and relates to the technical field of remote sensing image semantic segmentation. According to the method, a double-encoder and single-decoder structure, a multi-scale feature fusion channel and spatial path information are adopted; an adaptive efficient parameter fine tuning SAM2 large model knowledge is utilized to accelerate new task learning, global feature and local feature information of the SAM2 large model knowledge is effectively captured through a global local window multi-head self-attention module, a mixed Logit adjustment loss function based on mutual information is adopted to adjust category weight aiming at the problem of sample category imbalance of a traditional remote sensing data set, and the new task learning is accelerated. The remote sensing image semantic segmentation effect is improved; according to the method, pixel classification and ground feature correspondence are realized, the segmentation problems of complex background, unbalanced sample category and large scale change are effectively solved, the accuracy and stability of image segmentation are improved, and rapid and accurate segmentation of the remote sensing image is realized.

The invention discloses a medical image data processing method and system, and relates to medical image data processing, and the method comprises the steps: introducing an attention mechanism to calculate the attention weight of each mode based on the specific features of each mode, generating a specific feature map according to the attention weight, and carrying out the recognition of the specific feature map through a feature alignment algorithm, performing spatial alignment on the specific feature maps of the different modalities to generate an aligned specific feature map, and identifying correlation among the modalities by using a multi-head attention mechanism to generate a comprehensive feature map; and based on the comprehensive feature map, carrying out pixel classification through UNet to generate a preliminary segmentation result of the lesion area, meanwhile, utilizing an edge monitoring algorithm to obtain a multi-scale edge feature map of the lesion area, comparing the multi-scale edge feature map with the preliminary segmentation result, removing a pseudo lesion area according to a comparison result, and finally generating a lesion positioning map. According to the method, the pseudo focus area is eliminated through multi-scale edge feature comparison, and the accuracy and robustness of focus positioning are ensured.

In order to solve the problems that day lily plants are densely shielded and categories are difficult to distinguish, the invention provides a field dense day lily pixel classification and picking information acquisition method. The method comprises the steps of designing a double-branch decoder fusing local and global information, constructing a progressive stepped feature mining module (PLFM) and a multi-scale hierarchical category enhancement module (PCEN), optimizing hyper-parameters through a PALA algorithm, and constructing an adaptive double-branch loss enhancement and category enhancement network (DLCE-Net) to improve semantic segmentation performance. And designing a picking point positioning algorithm (MW-GCA) of Mini-Window guided angular point analysis based on a segmentation result, screening angular point coordinates through a U-shaped window, obtaining an attitude line segment, and realizing picking point pixel coordinate and angle estimation. The method can reduce the labor loss and accelerate the engineering landing of the automatic crop picking technology.

A process for tissue modelling from a medical image of a subject, for forming a three-dimensional (3D) model of a region of interest (ROI) of a subject with one or more tissue types, said process including the steps of (i) utilising a rule-based method that automatically generates the weak annotation, initial seed area from a medial image (210b); (ii) utilising a proposal generation method that integrates the multi-scale image features and anatomical prior (220b); and (iii) a comprehensive loss for CNN training that optimizes the pixel classification and feature distribution simultaneously (230).

The invention relates to a hyperspectral image classification method and device based on a state space model. The hyperspectral image classification method based on the state space model comprises the following steps: sequentially carrying out feature extraction and serialization processing on hyperspectral image data to obtain a shallow feature projection vector; performing global-local feature extraction on the shallow feature projection vector by adopting a neural network based on a state space model to obtain a fused feature projection vector; and carrying out pixel-by-pixel classification and dimension rearrangement on the hyperspectral image data in sequence to generate a classification result of the hyperspectral image data. According to the hyperspectral image classification method based on the state space model, long-range dependence modeling is achieved through the neural network based on the state space model with linear complexity, the calculation complexity is effectively reduced, and through feature fusion and residual error connection, the classification accuracy of the hyperspectral image is improved. And the perception capability of the neural network on different scale space-spectrum structures in the hyperspectral image is effectively enhanced.

The invention belongs to the technical field of terrain modeling, and particularly relates to a large-amplitude dynamic terrain generation method based on multi-source structure data, which comprises the steps of data sampling, pixel classification, waveband screening, spectrum fusion, feature fusion and model generation. According to the scheme, the data acquired by the optical sensor and the synthetic aperture radar sensor are respectively processed by using the double-flow type optimization U-Net model, so that each data flow can concentrate on the specific image characteristics of the respective sensor, and the requirement of vivid reproduction of a large-scale natural landscape is met; the method comprises the following steps: optimizing a ground mask by adopting morphological operation, firstly carrying out coarse classification on a topographic region to be measured, and then subdividing crop types of the ground surface according to a multispectral image and aperture radar data by adopting a method of combining plaque segmentation and feature fusion; therefore, the natural process of dynamically simulating the vegetation crops on the constructed digital terrain model along with the time change is realized.

The invention relates to the technical field of safety monitoring of machine vision, and particularly discloses a semantic segmentation-based high-altitude operation safety belt and safety rope detection method, which comprises the following steps of: performing polygonal labeling on figures, safety belts and safety ropes in a high-altitude operation picture by using a labeling tool, and distributing high-altitude operation category names; the method comprises the following steps: constructing a semantic segmentation model based on deep learning, carrying out convolution and down-sampling on an aerial work picture so as to obtain feature maps of different scales, using the multi-scale feature maps for predicting the size, position and category of masks, obtaining a prediction result, drawing the prediction result on the picture, and classifying all pixels in the picture by the masks of different categories; training the semantic segmentation model, and optimizing the parameter weight of the model; after training is completed, a semantic segmentation model is used for reasoning pictures, people, safety belts and safety ropes at different positions are distinguished, the method conforms to a current video monitoring system, and precision improvement is facilitated.

The invention discloses a laser beam positioning method based on a deep neural network. The method comprises the following steps: step 1, acquiring a light spot image of a laser beam on a reflector through image acquisition equipment; step 2, a deep neural network DeepLabV3 + model is utilized to perform pixel-by-pixel classification on the light spot image, light spot pixels and non-light-spot pixels are distinguished, a light spot mask is generated, the foreground pixel value of the light spot mask is set to be 1, and the background pixel value of the light spot mask is set to be 0; and step 3, performing morphological processing on the generated light spot mask, and calculating a centroid coordinate of the light spot by using a centroid method. The method has high noise immunity and generalization ability, can accurately and reliably extract the position and area of the light spot, and can give an accurate recognition result under the condition that the laser light spot is abnormal.

A rock breaking efficiency prediction method based on image recognition and machine learning comprises the following steps: acquiring a rock surface image, performing graying processing by applying an image processing technology, and then performing pixel classification and image segmentation so as to clearly divide different minerals into a plurality of independent structure diagrams; modeling a rock material according to mineral components, importing boundary information of the plurality of structure diagrams into a model, generating a plurality of large-particle filled rock blocks with the same size as the plurality of structure diagrams, and calibrating mesoscopic parameter values according to rock mechanical characteristics; establishing a hob rock breaking discrete element calculation model, and performing multiple intrusion rock breaking simulation tests on rock samples with different mineral contents; collecting simulation results to construct a rock breaking efficiency data set; a prediction model is established based on a PSO-BPNN algorithm, and a rock breaking efficiency prediction model is obtained after training; and carrying out real-time prediction on the rock breaking efficiency by utilizing the established model. According to the method, the influence of rock mineral components on the rock breaking effect of the hob is considered, and efficient and accurate prediction of the rock breaking efficiency of the hob is realized by utilizing a method of combining rock microscopic components and macroscopic rock breaking behaviors.

To correct a measurement dataset (15) from a lidar sensor system (3) comprising a detector array with a plurality of pixels, and where the measurement dataset (15) contains measurements that include an energy value and a distance value, a cluster (10) of pixels whose energy values ​​are greater than or equal to a threshold is identified. Based on the distance values, a distance value of the cluster (10) is determined, and pixels in an environment (11a, 11b) of the cluster (10) are identified as potentially faulty if their distance values ​​lie within a distance range around the distance value of the cluster (10). Depending on the energy value profile of the potentially faulty pixels, the potentially faulty pixels are classified as faulty, and their measurements are removed from the measurement dataset (15) or marked as faulty.

The invention discloses an SAR polarization calibration reference distributed target extraction method, and the method comprises the steps: carrying out the threshold method pixel classification through employing a polarization distortion parameter and a polarization orientation angle extracted from an uncalibrated full-polarization SAR image, and obtaining a pixel-level high-confidence volume scattering dominant target and a non-volume scattering dominant target; performing super-pixel segmentation on the uncalibrated full-polarization SAR image to obtain super-pixel units; extracting two polarization characteristics of the superpixel unit; processing the pixel-level high-confidence volume scattering dominant and non-volume scattering dominant targets by using a pixel proportion threshold method to obtain super-pixel unit high-confidence volume scattering dominant and non-volume scattering dominant targets, and combining the super-pixel unit high-confidence volume scattering dominant and non-volume scattering dominant targets with the two polarization features to obtain super-pixel unit positive and negative samples; and training the positive and negative samples of the super-pixel unit by using a random forest classifier to obtain a total image classification result, and the scattering dominant target of the super-pixel unit in the total image classification result being the reference distributed target meeting the polarization calibration requirement.

The invention relates to the technical field of remote sensing image processing, in particular to a hyperspectral image and laser radar image fusion classification method and device. According to the method, a deep network model is constructed and comprises a multi-level modal assimilation encoder and a multi-level assimilation modal reconstruction decoder; and for each hyperspectral pixel in the normalized hyperspectral image, inputting a neighborhood block of the current hyperspectral pixel and a neighborhood block of a laser radar pixel with the same position as the hyperspectral pixel into a multi-level modal assimilation encoder, and performing multi-stage modal information fusion and feature extraction to obtain the hyperspectral image. Obtaining a remote sensing assimilation mode of the current hyperspectral pixel and the corresponding laser radar pixel; and inputting each hyperspectral pixel and the remote sensing assimilation mode of the corresponding laser radar pixel into a multi-level assimilation mode reconstruction decoder to obtain a pixel classification result after each hyperspectral pixel and the corresponding laser radar pixel are fused. According to the method, the hyperspectral image and the laser radar image can be effectively fused and classified.

The invention relates to the field of image processing, and provides a seed multi-modal microscopic image acquisition method, device and equipment and a storage medium, and the method comprises the steps: carrying out the region segmentation of an obtained hyperspectral image of a plurality of seeds, and obtaining a single-seed region image; performing data extraction on the single seed area image to obtain spectral data of each seed; the spectral data comprises a seed identifier and a reflectivity value of each wave band; segmenting and extracting the acquired scanning images of the multiple seeds to obtain single seed scanning slices; performing three-dimensional reconstruction on the binary mask image to obtain seed three-dimensional physical structure information; the binary mask image is obtained by performing pixel classification processing on the single seed scanning slice. The method provides a basis for comprehensively and deeply analyzing chemical components and internal structures of the seeds, and improves the efficiency of seed detection and analysis.

The invention relates to a modeling method, in particular to an unmanned aerial vehicle spectral image BRDF rapid acquisition and adaptive generalization modeling method. The method comprises the following steps: 1, determining a research area, obtaining multi-angle BRDF data of the research area, and obtaining illumination change data at the same time; 2, preprocessing the acquired multi-angle BRDF data in combination with illumination change data to obtain preprocessed multi-angle images, splicing the multi-angle images to generate a panorama of the research area, and acquiring a digital surface model graph at the same time; 3, carrying out pixel-by-pixel classification on the research area based on the panorama to obtain a ground feature category map; calculating the gradient and the slope direction of the research area pixel by pixel based on the digital surface model graph; 4, correcting a preset BRDF model by using the acquired illumination change data, gradient and slope direction; and 5, self-adaptive generalization modeling is carried out. According to the method, high-efficiency BRDF acquisition can be realized, and the influence of regional change, space environment and topographic factors on BRDF modeling precision can be eliminated.

The invention provides a water body extraction method and device for a remote sensing image and electronic equipment, and relates to the technical field of remote sensing image data processing. The method comprises the following steps: performing LBP texture feature and gray feature extraction on a collected remote sensing image to obtain a pixel texture feature vector; inputting the remote sensing image into a water body pre-extraction model to obtain a preliminary water body extraction result, and constructing a water body extraction optimization model based on LBP-MRF according to the preliminary water body extraction result and the pixel texture feature vector; performing iterative optimization on a target energy function of the water body extraction optimization model, and updating a pixel classification label in the initial water body extraction result until the change of the target energy function meets a preset convergence condition; and determining a water body extraction image according to the updated initial water body extraction result. According to the technical scheme provided by the embodiment of the invention, information extraction of small water bodies and water bodies with complex edges in the remote sensing image can be better realized, and the accuracy and robustness of a water body extraction result are improved.

A computer apparatus and method for identifying and visualizing tumors in a histological image and measuring a tumor margin are provided. A CNN is used to classify pixels in the image according to whether they are determined to relate to nontumorous tissue, or one or more classes for tumorous tissue. Segmentation is carried out based on the CNN results to generate a mask that marks areas occupied by individual tumors. Summary statistics for each tumor are computed and supplied to a filter which edits the segmentation mask by filtering out tumors deemed to be insignificant. Optionally, the tumors that pass the filter may be ranked according to the summary statistics, for example in order of clinical relevance or by a sensible order of review for a pathologist. A visualization application can then display the histological image having regard to the segmentation mask, summary statistics and / or ranking. Tumor masses extracted by resection are painted with an ink to highlight its surface region. The CNN is trained to distinguish ink and no-ink tissue as well as tumor and no-tumor tissue. The CNN is applied to the histological image to generate an output image whose pixels are assigned to the tissue classes. Tumor margin status of the tissue section is determined by the presence or absence of tumor-and-ink classified pixels. Tumor margin involvement and tumor margin distance are determined by computing additional parameters based on classification-specified inter-pixel distance parameters.

The invention relates to a road extraction method and system based on semantic segmentation, and belongs to the technical field of remote sensing image processing, and the method comprises the steps: inputting a to-be-recognized remote sensing road image into a trained semantic segmentation model, and obtaining a first road segmentation result; extracting a road according to the first road segmentation result; wherein the semantic segmentation model performs reverse model parameter updating according to mixing loss during training; the mixed loss comprises pixel classification loss and connectivity loss; the connectivity loss is determined according to the predicted road center line and the real road center line; the predicted road center line is obtained by performing road center line extraction on a second road segmentation result, and the second road segmentation result is obtained by performing semantic segmentation on a sample remote sensing road image through a semantic segmentation model; the real road center line is obtained by performing road center line extraction on a real road segmentation result corresponding to the second road segmentation result. The method can improve the connectivity and integrity of the extracted road.

A farming machine includes one or more image sensors for capturing an image as the farming machine moves through the field. A control system accesses an image captured by the one or more sensors and identifies a distance value associated with each pixel of the image. The distance value corresponds to a distance between a point and an object that the pixel represents. The control system classifies pixels in the image as crop, plant, ground, etc. based on depth information in in the pixels. The control system generates a labelled point cloud using the labels and depth information, and identifies features about the crops, plants, ground, etc. in the point cloud. The control system generates treatment actions based on any of the depth information, visual information, point cloud, and feature values. The control system actuates a treatment mechanism based on the classified pixels.

The invention provides a low-gray-scale image imaging defect visual detection method and system for an OLED display screen. The method comprises the following steps: aiming at original image data of a display screen, extracting a gray scale picture with a pixel brightness mean value lower than a preset brightness threshold value as a first image, carrying out color characteristic analysis on the first image, and generating a second image marked with a chromaticity feature vector; identifying the chromaticity feature vector according to the second image, generating a feature vector matrix for representing chromaticity distribution of each pixel in the second image, and performing color jump analysis through the feature vector matrix to obtain a third image; extracting a jump region in the third image, and setting marks for distinguishing different dispersion distribution for chromaticity distribution dispersion distribution according to the jump region to obtain a fourth image; and identifying dispersion distribution according to the fourth image, combining with a chromaticity abnormal pixel region in the hopping region, performing pixel classification in the hopping region, and setting defect marks for pixels with imaging defects represented by a classification result to obtain a fifth image.

The invention discloses a vehicle-mounted track inspection system detection method, and relates to the technical field of tracks, an inspection starting point is set in a to-be-inspected line, a complete track image of the to-be-inspected line is pre-collected, a plurality of fastener images are generated through cutting with a fastener as the minimum unit, and track historical data are obtained; the track historical data are deployed on a cloud service side and a local inspection side at the same time; a vehicle-mounted track inspection system collects track real-time images, then the track real-time images are matched with an inspection starting point and cut into a plurality of fastener real-time images, region division is conducted on each fastener real-time image, track historical data of a local inspection side are called to conduct pixel grading comparison on different regions, and abnormal region images are screened; and incremental data is generated and transmitted to the cloud service side, track historical data is called to carry out refined detection on the abnormal region image, and a disease detection result is output. According to the invention, the problems of insufficient local detection performance, data accumulation and detection delay under high-speed operation of the electric passenger car are solved, and real-time track disease inspection is realized.

The invention discloses a submarine image splicing method based on suture line optimization, and the method comprises the steps: carrying out the dual-stage feature extraction of a target submarine image based on an improved ResNet50 network, and obtaining a multi-level feature map containing global structure information and local detail features; respectively carrying out global coarse registration and local fine correction on the second feature map and the first feature map based on a deformable convolutional network to obtain a sub-pixel level aligned deformed image; and carrying out cross-scale feature interaction and dynamic mask optimization on the deformed images based on a cross-image self-attention mechanism, dividing suture line weights between the spliced images, and carrying out pixel classification according to the weights to obtain seamless spliced images. According to the method, semantic fusion is guided through deformable convolution modeling elastic deformation and a self-attention mechanism, sub-pixel-level alignment and seamless splicing are achieved, and the problems of geometric dislocation and suture line artifacts caused by refraction distortion, scattering noise and dynamic deformation in large-parallax submarine image splicing are solved.

The invention discloses an aging detection method, device and equipment for transformer oil and a medium, and the method comprises the steps: obtaining offline detection data and oil visual data corresponding to a plurality of aging stages of the transformer oil by responding to a modeling request; carrying out pixel classification and labeling on the oil visual data to obtain a classification labeling image; training a preset feature segmentation model by using the offline detection data and the associated feature pair of the classification annotation image to obtain a target feature segmentation model; when the sample data and the sample image of the to-be-detected transformer oil are received, the target feature segmentation model is called to be combined with the quantitative weight, and the aging grade of the to-be-detected transformer oil is determined. Therefore, the machine learning method is applied to realize the identification of the aging characteristics of the transformer oil and the judgment of the aging grade, and the aging characteristics can be identified and judged more comprehensively and more accurately in a mode of combining the aging characteristic detection indexes and the visual images.

Systems and methods for capturing and processing image data captured by an optical microscope, such as an Open Top Light Sheet (OTLS) microscope are disclosed. Image data can be processed by techniques including flat fielding, image depth correction, and edge correction, and pixel classification and image segmentation techniques can be applied to more accurately identify lipid droplets and fibrous structures for assessment of, for example, steatosis and fibrosis in human liver biopsy tissue samples. Systems and methods are also related to capturing low resolution images of a sample, identifying regions of interest, and then capturing high resolution image of the regions of interest to reduce memory used for storage, increase computation speed, and reduce computational power.

The invention discloses an ultrasonic phased array weld defect intelligent identification method based on adaptive filtering, and belongs to industrial nondestructive detection image identification methods, the technical scheme is that the method comprises the following steps: sampling and sorting pipeline weld PAUT detection images; dividing the collected image samples into a training set and a test set; pixel classification and filtering noise reduction are carried out on the training set images through a hybrid adaptive filter; performing abnormal signal labeling on the preprocessed training set image and the unprocessed test set image respectively; training a target recognition network through the preprocessed training set; and collecting a new fan-shaped scanning image of the pipeline weld defect as network input, and identifying the weld defect. The ultrasonic phased array weld defect intelligent identification method based on self-adaptive filtering has the beneficial effect that the ultrasonic phased array weld defect intelligent identification method based on self-adaptive filtering is provided.

The embodiment of the invention discloses a water body information extraction method based on a remote sensing image, and relates to the technical field of remote sensing image processing, and the method comprises the steps: obtaining a green light wave band value, a red light wave band value and a near-infrared wave band value of a target pixel in a to-be-processed remote sensing image; based on the green light wave band value, the red light wave band value and the near-infrared wave band value, calculating a water and land pixel measurement variable of the target pixel through a preset water body index formula; comparing the land and water pixel metric variable obtained by calculation with a preset threshold value, and classifying the target pixel as a water pixel or a non-water pixel based on a comparison result; the accuracy of land and water pixel measurement can be further improved, the phenomenon that the water body area is confused with other interference areas such as roads, building roofs and bridges is reduced, and water body recognition is more accurate.

The invention discloses a semantic segmentation method and device based on a lightweight coding-decoding network, equipment and a medium. The method comprises the following steps: acquiring image data; preprocessing the acquired image data; inputting the preprocessed image into an encoder for low-resolution feature extraction; the extracted low-resolution features are input into a decoder for feature detail recovery, and a feature map with the same resolution as the original image is obtained; and inputting the feature map with the same resolution as the original image into a semantic segmentation classifier to obtain pixel-by-pixel classification prediction. Therefore, the method can achieve higher class precision and more detailed object contour prediction, thereby achieving better balance between speed and precision.

The invention belongs to the technical field of machine learning, and particularly discloses a plateau wetland landscape component extraction method and system and a program product, and the method comprises the steps: designing a spectrum-space double-branch interaction model, extracting the spectral features of pixels at a spectrum branch through a pixel dimension raising module and a ViT architecture Transformer encoder, and carrying out the extraction of the components of the plateau wetland landscape component; extracting neighborhood spatial features of pixels through a multi-scale feature extraction module and feature enhancement in a spatial branch, then integrating the features of the two branches through a cross attention fusion module, outputting pixel classification results, and determining each plateau wetland landscape component region in the remote sensing image according to each pixel classification result. The high-efficiency and reliable extraction of the plateau wetland landscape components can be realized. The plateau wetland landscape component classification method can effectively relieve the problem of salt and pepper noise existing in a traditional pixel-by-pixel classification method, guarantees the light weight of the network while improving the plateau wetland landscape component classification precision, and provides more accurate technical support for long-time-sequence plateau wetland landscape component extraction.

The invention relates to a target detection-based iris rapid and accurate segmentation method, which comprises the following steps of: obtaining an eye image for training, establishing a three-dimensional coordinate system and obtaining parameters of an iris ellipse by utilizing a three-dimensional mathematical relationship; constructing a directional bounding box fitted with the contour of the iris ellipse, the four edges of the directional bounding box being tangent to the iris ellipse, and solving the coordinates of the four vertexes of the directional bounding box; carrying out model training by taking the coordinates of the four vertexes of the directional bounding box and the parameters of the iris ellipse as model training labels so as to obtain a network model; the network model is used for detecting a directional bounding box surrounding an iris ellipse of the target eye image, and calculating and segmenting the iris ellipse according to the directional bounding box surrounding the iris. According to the method, a traditional iris segmentation mode depending on edge detection or pixel classification is abandoned, a brand new technical route based on deep learning target detection is adopted, and the limitation of a traditional method in processing a low-quality iris image is effectively overcome.

The invention discloses a two-stage parameter adaptive optimization honeycomb lung focus segmentation method, a storage medium and equipment, and relates to the technical field of honeycomb lung image processing methods. The method comprises the following steps: image preprocessing: acquiring a honeycomb lung CT image, preprocessing the CT image, and optimizing a preprocessing process of the CT image by adopting a mean shift algorithm; coarse segmentation: performing lung coarse segmentation on the preprocessed image by using a classification method based on an SNIC algorithm and a random forest and a superpixel classification method; and fine segmentation: performing fine segmentation on the coarse segmentation result and the original CT image based on an adaptive optimization network SRU-Net. The method can improve the perception capability of the model for focus boundary details, and improve the segmentation precision and stability.

Identifying hazard trees is described. An example method includes receiving multiple images for a geographic area that includes multiple electrical assets of a power distribution infrastructure. Pixels of the multiple images are classified as hazard tree pixels or as non-hazard tree pixels using one or more convolutional neural networks. Multiple polygons are generated based on the hazard tree pixels, a polygon corresponding to one or more hazard trees in the geographic area. A height of a polygon and a distance from the polygon to an electrical asset is determined. Based on the height and the distance, the one or more hazard trees corresponding polygon are determined to represent a potential hazard to the electrical asset. A notification of the potential hazard to the electrical asset is generated and provided.