37 results about "Automatic threshold" patented technology

The invention relates to an unsupervised anomaly detection method based on an enhanced variational auto-encoder, and the method comprises the following steps: S1, data collection and preprocessing: obtaining a multi-dimensional airborne pollutant monitoring data set, and carrying out the filtering and denoising, and missing value interpolation data cleaning; s2, model construction: two layers of TCN residual blocks are adopted, each residual block internally comprises two layers of one-dimensional expansion causal convolution, and a TCN-based encoder totally comprises four layers of expansion causal convolution; the method comprises the following steps: constructing a full connection layer behind an encoder output layer based on TCN, and mapping output feature dimensions into a two-dimensional submerged space; the VAE decoder module uses two layers of full connection layers, and outputs a feature dimension as an input data dimension; s3, constructing an exception score calculation function and an automatic threshold exception detection function: calculating an exception score of each sample based on a reconstruction loss calculation method, setting an exception threshold, and identifying an exception sample; and S4, model training.

The invention relates to the technical field of industrial nondestructive testing, in particular to a wall thickness detection device and method based on a dual-ray system, and the method comprises the steps: synchronously shooting vertical and horizontal projection images of a pipe through a dual-ray source; extracting first line gray data of the bidirectional projection image, and capturing an edge position through an automatic threshold method; grabbing inflection points of the inner wall and the outer wall of the pipe according to the edge position, and totally obtaining eight inflection points; calculating inner and outer diameters of the pipe by using a mathematical model based on the inflection point position Calculating the wall thickness, the ovality, the eccentricity value and the concentricity according to the inner diameter and the outer diameter; according to the invention, full detection in the pipe production process is realized, the detection efficiency is high, the influence of slight shaking of the pipe can be resisted, the detection result is accurate, and the device is suitable for detecting the wall thickness and related quality indexes of linear materials such as PVC pipes and cables.

The invention relates to the technical field of radio frequency chip performance testing, and particularly provides a real-time visualization method and visualization system for radio frequency chip performance testing, traditional single-thread sequence control is replaced by a multi-thread asynchronous acquisition architecture, and the testing efficiency is remarkably improved; through adding a timestamp label mechanism to each piece of test data, the cross-device data synchronization precision reaches a microsecond level, and the problem of asynchronous triggering of a radio frequency signal and direct current bias is solved. The radio frequency index data table constructed by automatic association realizes intelligent aggregation analysis of multiple parameters under multiple combinations, and avoids manual matching errors caused by data dispersed storage in a traditional scheme. Based on an automatic threshold setting and real-time judgment function of a test protocol, intelligent switching of judgment standards under a multi-protocol standard is supported, errors caused by manual configuration are reduced, and modification during protocol updating is avoided. And real-time visual rendering enables a user to dynamically observe a multi-parameter association change trend, so that the problem positioning efficiency is improved.

The invention relates to the technical field of intelligent design of optical lenses, and discloses a multi-series superposed complex periodic structure arrangement method and system.The method comprises the steps of normalized coordinate transformation, multiple sine wave superposition, an iterative function system, a Julia set algorithm, an LSystem grammar system, parameter adjustment, moire fringe fusion and the like; and a complex periodic structure with high complexity, adaptability resistance and natural morphological characteristics is generated. Advanced technologies such as logarithm uniform distribution wave number generation, a chaos game algorithm, linear mapping complex number conversion, a turn graph interpreter, multi-scale entropy complexity quantification and an Otsu automatic threshold method are adopted, and accurate parameterization control and manufacturing adaptation are achieved. The system comprises a coordinate transformation module, a wave interference calculation module, a fractal generation module, a complex power system module, an LSystem processing module, a parameter control module and a contour surface extraction module. According to the invention, a systematic solution can be provided for complex structure design of optical devices such as myopia prevention and control lenses.

The invention discloses a brain tumor real-time identification system and method based on cortical electroencephalogram signals, and relates to the technical field of medical artificial intelligence, and the method comprises the following steps: obtaining multi-channel ECoG signals in real time; the multi-channel ECoG signals are updated in real time, the multi-channel ECoG signals updated in real time are preprocessed, and the preprocessed multi-channel ECoG signals are obtained; performing channel feature extraction on the ECoG signal of each channel to obtain various ECoG signal features under each channel as channel features; calculating segmentation thresholds of different channel characteristics under each channel based on an automatic threshold segmentation algorithm; carrying out confidence evaluation on the segmentation threshold values of different channel characteristics under each channel; and marking the channel of which the confidence coefficient is smaller than a set tumor marking threshold value as a tumor area. Accurate positioning of a tumor area can be achieved, functional state changes of the cerebral cortex can be reflected in real time, and doctors can conveniently master real-time changes of cortex functions in time.

The invention relates to a three-dimensional modeling method for a reef limestone microstructure, which comprises the following steps of: scanning to obtain an original grayscale image of a rock core sample, and preprocessing the original grayscale image; based on the preprocessed grayscale image, extracting a sample shell mask; establishing a Gaussian mixture model of sample gray level distribution, and determining a segmentation threshold value; classifying and segmenting the volume data through the segmentation threshold to obtain a pore candidate region, a matrix candidate region and a pending region; performing morphological closed operation on the candidate region to obtain a pixel set belonging to the pore and a pixel set belonging to the matrix; and classifying and segmenting the pending region by adopting an energy minimization algorithm based on graph segmentation. According to the method, the GMM dual-threshold segmentation and shell mask construction technology is adopted to accurately extract the core area, and air at the core boundary is prevented from being wrongly segmented into pores, so that the real boundary of the core is completely reserved; according to the method, the pores can be finely judged, the automatic threshold segmentation result is stable and reliable, and manual intervention is greatly reduced.

The present invention discloses a method for rapid water body extraction from remote sensing images using multi-scale GLCM feature fusion, which relates to the technical field of water body remote sensing monitoring. The method comprises the following steps: S1, data collection; S2, calculating a water body index matrix W based on the collected data; S3, extracting multi-scale GLCM texture features; S4, weightedly fusing the water body index matrix W with multi-scale texture features to generate a comprehensive feature matrix; S5, defining a parameter space; S6, automatic threshold segmentation using the Otsu method; S7, generating a binary water body matrix; and S8, iteration. The present invention has the beneficial effect of introducing a Monte Carlo sampling and iterative optimization mechanism, performing a global search in a broad parameter space, finding an optimal parameter set, achieving fine weighted fusion of features at different scales, and automatically optimizing the weighted parameters of texture features, thereby significantly improving the accuracy and efficiency of water body recognition in complex environments, while reducing the need for manual parameter adjustment and having higher applicability.

The application discloses a KPI anomaly detection method fusing Bi-TCN and an adversarial VAE. The method comprises the following steps: S1, data preprocessing; S2, model initialization, random initialization and assignment of model parameters, and parameter setting; S3, model training; and S4, model testing, same data processing is performed on the test, a reconstructed error is obtained by inputting the trained model, and a threshold value generated by using an automatic threshold method is used to determine whether an anomaly exists. The application uses a variational autoencoder for adversarial training, the VAE is more robust to noise and abnormal values, and adversarial training can more efficiently amplify the reconstructed error of an input containing an anomaly to distinguish normal KPI data from abnormal KPI data.

The present invention relates to a system for detecting and quantifying the volume of underwater seaweed based on multibeam water column data, a method thereof, and a recording medium thereof. More specifically, the invention relates to a system for detecting and quantifying the volume of underwater seaweed based on multibeam water column data that visualizes water column data acquired by a Multibeam Echo Sounder (MBES) in real time and quantifies the spatial distribution and volume of seabed seaweed through a user interaction-based detection workflow, as well as a recording medium thereof. The underwater seaweed detection and volume quantification system based on multi-beam water column data according to the present invention is a system for processing water column HDF5 data acquired by a multi-beam echo sounder (MBES), comprising: a data management module that loads amplitude, transverse coordinate (X), depth coordinate (Z), UTM coordinate, heading, and ping number from water column data in HDF5 format, and converts transverse-depth coordinates (X, Z) into UTM absolute coordinates through rotational transformation using heading; and a Fan Plot visualization module that includes an adaptive grid rendering function that converts water column data of each ping loaded in the data management module into a 2D grid of a transverse-depth plane, renders it in real-time as a Fan Plot, and dynamically adjusts the grid resolution according to the viewport magnification ratio, and an ROI editing function that generates an arbitrary polygonal ROI by user mouse input and automatically propagates the polygonal ROI to adjacent pings when the ping moves. The system is characterized by comprising: a multi-stage filtering pipeline module that sequentially performs amplitude range filtering, Ray-casting vectorization-based Point-in-Polygon discrimination, intensity percentile filtering, Otsu automatic threshold determination, morphological closing operation, and DBSCAN clustering on the order data loaded into the data management module; a 3D analysis module that performs KDTree-based nearest neighbor matching between sets of filtered result points of adjacent pings in the order data filtered by the pipeline module, generates a 3D point cloud through linear interpolation, and estimates volume through uniform voxel grid transformation; and an output module that outputs the filtering points, interpolated point cloud, voxel mesh, and volume statistics of the pipeline module and the analysis module in CSV, PLY, and OBJ formats.

The invention belongs to the technical field of vehicle exhaust emission monitoring, and discloses a heavy diesel vehicle emission degradation prediction system and method. The method comprises the following steps: vehicle type family classification and PEMS data source acquisition; pEMS data transient acquisition and steady-state window extraction are carried out; building and training a multi-mileage model; based on the accuracy verification result of the multi-mileage model, creating a digital twinborn body for each in-use vehicle in the vehicle type family; calculating a preliminary degradation coefficient; and calculating a correction degradation coefficient according to the predicted correction value of the reference model and the predicted correction value of each current mileage section model, comparing the correction degradation coefficient with a preset degradation threshold value, judging whether the correction degradation coefficient exceeds the preset degradation threshold value or not, if yes, carrying out preset degradation threshold value early warning, and carrying out in-vehicle maintenance closed loop according to early warning information. According to the invention, real-time accurate monitoring is realized, and low-cost, real-time and full-coverage emission degradation monitoring is realized; and automatic threshold judgment and precise maintenance closed loop are realized, and the supervision efficiency and compliance rate are improved.

The application discloses a power operation safety evaluation system based on big data, which comprises a safety evaluation data acquisition module, a multi-dimensional safety feature selection module, a dynamic safety evaluation model construction module and an intelligent safety evaluation module. The application relates to the technical field of power safety data processing, and specifically relates to a power operation safety evaluation system based on big data. The application innovatively introduces a self-encoder model, accurately quantifies the nonlinear correlation between features by combining cosine similarity, determines an optimal sparse correlation graph through an automatic threshold adjustment algorithm, generates feature clusters by algorithm clustering, automatically identifies the optimal clustering structure, filters out double-optimal core features in terms of applicability and centrality, significantly reduces redundant features, enhances the output result accuracy of the evaluation model, and constructs a dynamic safety evaluation model based on the multi-view construction of feature clusters and the fusion of a double-cycle safety evaluation unit, thereby significantly improving the accuracy and stability of the power operation safety evaluation model.

The application provides a computer vision-based structural vibration displacement monitoring method and system and a storage medium, and the method comprises the following steps: step 1, structural vibration video preprocessing; basic information of video data is automatically acquired by using a video basic information automatic acquisition algorithm, and an image gray scale and local histogram equalization processing method is used to simplify the image calculation complexity and enhance the image features; step 2, structural displacement monitoring based on improved FAST corner detection; an automatic threshold calculation method based on the maximum inter-class variance is introduced to form an improved FAST corner detection algorithm, the image features obtained by the improved FAST corner detection algorithm are combined with the pyramid Lucas-Kanade optical flow method, and a whole-process automatic image feature detection and motion tracking scheme is established. The method has the advantages that the overall monitoring process is optimized, and the method is non-contact, simple to operate and high in automation.

The application discloses a four-wave transverse shearing interferogram effective area segmentation system and method based on FFT and global threshold, and belongs to the technical field of optical interference measurement and wavefront sensing. The method comprises an image acquisition module, a preprocessing module, an FFT low-frequency extraction module, an automatic threshold segmentation module and a morphological optimization module. The application carries out smooth processing on the background brightness distribution of the interferogram through FFT low-pass filtering, and realizes automatic segmentation of the effective area in combination with Otsu global threshold, effectively avoids technical defects such as center area inversion blackening, edge fracture and excessive noise that are prone to occur in traditional area segmentation methods, can completely extract the four-wave overlapping area, has the advantages of stable segmentation, strong anti-interference capability, no need for manual parameter adjustment, good real-time performance and the like, and can be widely applied to various four-wave transverse shearing interferometer wavefront detection systems.

The application discloses a pretreatment method suitable for liquid-phase plasma discharge spark image recognition, and specifically comprises the following steps: step 1, an industrial camera is used to take pictures of the surface of a workpiece to obtain a to-be-processed spark image; step 2, the spark image obtained in step 1 is subjected to grayscale processing to obtain a grayscale histogram; step 3, a background correction method based on a rolling ball algorithm is used to correct the background of the to-be-processed spark image subjected to grayscale processing in step 2; and step 4, an optimal segmentation threshold is obtained by using an automatic threshold segmentation method based on a grayscale mean value and a grayscale standard deviation, the spark and the background in the spark image processed in step 3 are segmented, and a target spark image is obtained. The application solves the problem that the existing threshold segmentation method cannot accurately segment small target discharge sparks in a complex background with high grayscale levels, uneven image brightness and a multi-peak grayscale histogram.

A system and method for predictive maintenance of operational mechanisms using dual-stage artificial intelligence analysis of optical sensor data. The system comprises optical sensors capturing images, a hardware interface, and a processor implementing two specialized AI models: a first model generating structured technical captions describing physical conditions, and a second model identifying mechanical anomalies from these captions. The system features dynamic control parameter modification, automatically implementing adjusted operational parameters based on detected anomalies until maintenance completion. Real-time severity assessment enables quantitative evaluation with automatic threshold modifications to prevent mechanical failure. The system generates comprehensive maintenance alerts including technical specifications, sensor data, time-to-failure predictions, and maintenance procedures. Implementation options include local, cloud-based, or hybrid processing configurations. Applications span critical transportation systems, defense platforms, and industrial equipment, providing continuous monitoring, predictive maintenance, and automated control adaptation to prevent equipment failure.

The application relates to an FPGA-based implantable electrode Spike signal detection system and method, which uses FPGA to collect brain electrical data signals transmitted by an implantable electrode chip, carries out filtering processing and automatic threshold feature extraction on the collected brain electrical data, completes the Spike extraction task, and sends the processed brain electrical data and the original data to an upper computer through a PCI-E mode. Compared with the prior art, the application adopts FPGA parallel processing data, integrates the Spike signal detection method on the FPGA, greatly reduces the real-time detection speed of the Spike, adopts a filter to pre-process the brain electrical data, can effectively retain Spike effective signals, adopts an automatic threshold feature extraction method, can adaptively extract Spike signals according to brain electrical signal fluctuations, not only reduces the power consumption of the chip, but also guarantees the accuracy of Spike signal extraction.

The application provides a kind of based on scanning electron microscope image deep-sea sediment particle three-dimensional structure reconstruction method: obtain the scanning electron microscope image of deep-sea sediment particle and carry out gray scale, filtering and contrast enhancement;Based on automatic threshold segmentation and connected domain analysis lock target particle area, and synchronous cutting gray scale and binary image;To target particle binary mask is maximum connected domain reservation, hole filling and size standardization processing;Gray value is mapped to relative height and shield background area, generate only cover target particle height field;Based on the height field generates multi-view three-dimensional graph, color height distribution graph, center profile, error band, local refinement reconstruction graph and height classification graph.Through the technical scheme of the application, the efficiency and stability of particle morphology contrast analysis can be improved without changing the existing scanning electron microscope hardware, realizing the automatic extraction, background shielding, three-dimensional morphology reconstruction and quantitative expression of deep-sea sediment particles in a single SEM image.

The present invention discloses a SAR image adaptive threshold segmentation method and device, which relate to the technical field of remote sensing image processing. The present invention first extracts a larger target area in the SAR image through coarse threshold segmentation, automatically selects background points according to the target in the image and extracts a profile line from the target center to the background point, calculates the threshold corresponding to the target based on the profile line, adaptively divides the image into blocks according to the target, and automatically selects a local threshold to achieve adaptive threshold segmentation of the SAR image; compared with the existing adaptive threshold segmentation method and the semi-automatic global threshold segmentation method, the present invention adaptively divides the image into blocks with the large target as the center, automatically draws a profile line across the target and the background for the block image, and automatically extracts the local threshold, thereby overcoming the problems of regular grid division, automatic threshold extraction, manual line drawing, etc. in the existing methods, greatly improving the effect of SAR image threshold segmentation, and can be applied to water body extraction, shadow detection, bright target detection, etc.

The invention discloses a battery pack EOL test data processing and judging method, system and device and a medium. The method comprises the following steps: acquiring basic information and configuring a test mode; automatically identifying and positioning an original data workbook; carrying out secondary calibration on the number of multi-packet combination packets and dividing data column segments; accurately positioning the working steps and extracting key data; dCIR key indexes are automatically calculated and summarized; carrying out dynamic test automatic identification and typing calculation; automatic threshold value judgment and abnormal popup alarm are carried out; and automatic filing, anti-overlay storage and MES synchronous uploading are realized. The method can adapt to the existing data generation rule of the test equipment, reduces the manual dependence, and improves the consistency and traceability.

The present invention proposes an automatic threshold color factor algorithm for green vegetation image segmentation, which is based on the automatic threshold color factor of normalized green channel (g channel) information. The algorithm consists of two parts. First, based on ExG in the color factor and the automatic threshold algorithm Otsu, the rough vegetation position information in the image is obtained to filter the feature pixel rows. Then, the normalized g channel information is extracted from the filtered feature pixel rows, and the threshold is automatically determined according to its change law. Subsequently, the g channel of the entire image is binarized using the determined threshold, and filtered and hole-filled to form a mask image. Finally, the mask image is used to perform an AND operation with the original color image to achieve the segmentation of the vegetation area. This method is used to conduct an experimental comparison with five color factors combined with the Otsu algorithm. The experimental results show that this method can quickly and stably achieve automatic segmentation of green vegetation in the image.

The invention provides an electron microscope image splicing method, which comprises the following steps of: calculating initial splicing positions of a plurality of images to be spliced, and obtaining overlapping areas among the images at the initial splicing positions; calculating the relative offset of the two images in the overlapped region by adopting phase correlation; the relative offsets are screened through confidence coefficients, and the relative offsets with the confidence coefficients larger than an automatic threshold value are reserved as effective position relative offsets, so that the smoothness of adjacent image contents is guaranteed; grouping the images according to the connectivity, calculating the accurate position of each image in each group, performing intra-group adjustment, calculating the relative position between different groups, and performing inter-group adjustment; and fusing the adjusted images. By combining the parameters of the images shot by the electron microscope and the image phase correlation algorithm, the complex transformation between the images can be processed, and the whole image can be quickly and accurately spliced.

The application discloses a detection device and method for positive and reverse surfaces of an interference type ultra-small curvature plano-convex lens or plano-concave lens. The detection method comprises the following steps: obtaining an interference image of a surface of a sample to be detected; and determining whether the surface is a curved surface or a flat surface according to the shape of the interference image. The detection method is simple, easy to operate, high in efficiency and 100% in accuracy. The detection device for positive and reverse surfaces of the interference type ultra-small curvature plano-convex lens or plano-concave lens is simple in structure and convenient to use. After the optical system with the specific structure is used, the interference image of the surface of the lens can be obtained by using a plane array camera, and whether the surface is a curved surface or a flat surface can be determined according to the shape of the obtained interference image. Further, after the interference image is sequentially subjected to Fourier transform and automatic threshold value calculation related to a mode, the difference between the calculation results of the curved surface and the flat surface can be used to realize automatic reading of the curved surface and the flat surface.

Embodiments of digital polymerase chain reaction (dPCR) systems and methods with improved automatic threshold performance and accuracy are disclosed. One embodiment of the present disclosure includes a dPCR processing module that includes an automatic threshold proxy that dynamically analyzes and selects between Gaussian mixture model (GMM) and K-means clustering for automatic threshold results of dPCR assays using a dPCR instrument to improve dPCR measurement techniques.

The invention belongs to the field of cornu cervi pantotrichum decoction piece identification, and particularly relates to a cornu cervi pantotrichum decoction piece identification method based on image identification. The system comprises a pilose antler decoction piece image feature extraction module, a pilose antler decoction piece image pore identification module and a pore statistics output module. Through an OpenCV technology, preprocessing and pore feature extraction are carried out on a cornu cervi pantotrichum decoction piece image, a quantitative index system of pore size, number and distribution density is constructed, a grading standard based on a pore feature threshold is established, and through automatic threshold determination, cornu cervi pantotrichum decoction piece image data can be processed in batches and a standardized grading result can be output. Compared with traditional artificial experience judgment, the method is suitable for quality evaluation and grade division in industrial production. Particularly, the problem that the classification accuracy is not high due to the fact that the characteristics of the transition area samples of the cornu cervi pantotrichum decoction pieces are not obvious is solved through the method of visualization and quantification of the transition area samples of the cornu cervi pantotrichum decoction pieces. Furthermore, the standardized, large-scale and featured development of the deer industry is promoted, and the standardization of the pilose antler traditional Chinese medicine is promoted.

The invention provides a retina layered line identification method based on an OCT image. The method comprises the following steps: acquiring the OCT image and a corresponding laser fundus image; performing automatic threshold segmentation on the OCT image, and identifying a vitreous body-nerve fiber layer layering line in combination with an image gradient and shortest path algorithm; identifying an optic disk area of the OCT image through an image registration technology, and dividing the OCT image into an optic disk area and a non-optic disk area; adopting a convex hull calculation method to generate a retina layering line of the optic disc area; based on the vitreous body-nerve fiber layer layering line, segmenting a retina layering line of a non-optic disc area layer by layer through gradient weight distribution and a shortest path algorithm; and integrating the layered lines of the optic disc region and the non-optic disc region, and outputting a complete retinal layered structure. According to the technical scheme, the accuracy and automation degree of retina layering are improved, optic disc and non-optic disc areas are considered, layering is complete and coherent, and the method is suitable for clinical auxiliary diagnosis and subsequent analysis.

The invention discloses a character stroke disconnection defect detection method and device, electronic equipment and a medium, and relates to the technical field of image recognition. The method comprises the following steps: acquiring a real-time character image and a template character image; performing global automatic threshold segmentation on the real-time character image to obtain a first binary image, and performing defect detection operation on the first binary image to obtain a few-print defect detection result of the first binary image; performing secondary binarization processing based on a character region on the real-time character image to obtain a second binary image, and performing defect detection operation on the second binary image to obtain a damage defect detection result of the second binary image; and based on the underprint defect detection result and the damage defect detection result, executing overlapping judgment, and outputting a stroke disconnection defect detection result of the real-time character image. With the adoption of the method, the detection precision can be ensured, the tiny disconnection defects of character strokes can be accurately identified, and the reliability and precision of character defect detection are improved.

This invention belongs to the field of intelligent cooking control technology and discloses an intelligent detection method for the cooked doneness of pan-fried basa fish based on multimodal data fusion. The steps are as follows: A sample for creating a cooked doneness classification model of pan-fried basa fish is prepared. Hyperspectral data, infrared thermal imaging data, and electronic nose data of the pan-frying process are collected simultaneously, and automatic threshold segmentation and filtering denoising preprocessing are performed on the multimodal data respectively. Multi-branch convolutional feature extraction modules are designed for hyperspectral, infrared image, and electronic nose signals respectively. A Bayesian-optimized hyperparameter automatic search strategy is designed to improve model performance. Relative position encoding and sparse attention mechanism are introduced into the Transformer network structure to construct a basa fish cooked doneness MBCE-BO-RP-Transformer classification model. The classification accuracy and lightweight performance of the model are tested and evaluated.

The invention discloses a collision protection system for precision machining equipment, which comprises data acquisition and automatic threshold setting, and is characterized in that an intelligent collision shield system comprises the following steps: step 1, intelligent collision detection, step 2, data analysis and prediction, step 3, real-time alarm and shutdown, and step 4, remote monitoring and maintenance and vibration acquisition judgment. The vibration condition of the spindle is monitored in real time, an alarm is automatically given when the machine tool is abnormal in collision, the spindle precision is protected through cutter breakage, and shutdown control is provided when collision occurs. According to the collision protection system of the precision machining equipment, through accurate data collection and intelligent analysis, cooperation with rapid protection execution, completion of a data flow closed loop and providing of all-directional collision protection, tiny signs such as abnormal dynamic states and collision risks can be accurately captured, and according to an intelligent analysis result, preset protection measures are started, emergency shutdown is executed, and protection benefits are presented; the collision loss rate is obviously reduced, the downtime is shortened, the maintenance cost is reduced, and the operation safety is improved.

The application discloses a kind of spot defect detection methods of airborne optical fiber end face, first to optical fiber image acquisition is carried out, then grey levelization obtains image Img1, then binaryzation is carried out.Edge detection is carried out to image, the outer edge contour of optical fiber is extracted, and the outer edge contour extracted is used as ROI region of interest, and the grey image in the region is operated with mask, and image Img2 is obtained.Automatic threshold binaryzation processing is carried out to image Img2, and image Img2 is carried out binaryzation segmentation;Automatic extraction is carried out to highlight spot in image Img2, to detect the defect area S' of highlight defect part of optical fiber.The ratio F of defect area S' and the area S of image Img2 is used as defect rate.The application can accurately extract highlight defect part, realizes automatic defect detection, with the advantages of high efficiency, simple, stable and reliable.

The application discloses a kind of mixed intelligent driving three-dimensional point cloud change detection method, comprising the following steps: double time phase point cloud input and pre-processing;Cross-period corresponding relationship between double time phase point cloud is established using neighborhood search;Multi-scale normal distance geometry prior calculation;Texture difference prior calculation;Automatic threshold and normalization processing are carried out to the above calculation result;Density consistency suppression and prior score construction;Change perception difference attention coupling is carried out to the double branch coding-decoding network constructed and prior knowledge injection;Change perception enhancement features are obtained;Decoding output change detection result.The application significantly enhances the attention and expression to the real change area without changing the structure of the main network, effectively reduces the false detection caused by density difference, occlusion and noise, and realizes high-precision detection of the real change area in double time phase urban three-dimensional point cloud, and improves the adaptability and robustness of the method under complex occlusion, inconsistent density and multi-source data conditions.