43results about How to "Guaranteed segmentation effect" patented technology

The present invention discloses a method for detecting and locating insulator defects in power transmission line image via a drone, and belongs to the technical field of image extraction and recognition. The detection and location method for the insulator defects comprises the steps of: S1. converting an aerial image from an RGB color space into an HSI (Hue-Saturation-Lightness) space, and carrying out binarization on extracted H component image and S component image separately to obtain a binary image of a preliminary contour of an insulator string; S2. using an ant colony algorithm based on a particle swarm optimization parameter to extract a contour of an insulator monomer; and S3. using a least square method to perform ellipse fitting on the contour of the insulator monomer, and locating the insulator defects by detecting the space between the contours of insulator monomers. By using the technical scheme provided by the invention, the detection precision of the insulator defects can be significantly improved, background interferences are reduced, and the operation speed is relatively high.

The invention discloses an iris image segmenting method and device. The method includes performing rough positioning on a pupil in an iris image; expanding an image in a candidate area on the outer boundary of an annular iris into a rectangular image in a polar coordinate system and acquiring an excircle boundary estimation value of a target iris according to the gray level change of the rectangular image; intercepting an upper eyelid zone and a lower eyelid zone of the iris image according to the excircle boundary, acquiring boundary points of an upper eyelid and a lower eyelid, performing curve fitting on the boundary points of the upper eyelid and the lower eyelid, and obtaining an upper eyelid boundary curve and a lower eyelid boundary curve; performing noise detection on valid zones determined according to the upper eyelid boundary and a lower eyelid boundary; performing fine positioning on the excircle boundary of the iris through a radial projection method and performing fine positioning on the boundary of the pupil by utilizing a calculus detection operator. The invention solves a problem of low accuracy of low quality iris image segmentation performed by iris image segmenting methods in the prior art.

The invention provides a method for manufacturing femtosecond laser complex components in a large batch. The surface processing of the complex component comprises: setting, according to the area and the focal depth of a laser processing region, a constraint condition and a region growing algorithm by using a complex component model reconstruction and surface processing pattern splicing algorithm;dividing the processing region, simplifying the region boundary to obtain the processing region, extracting the key features of the processing pattern types and layout features; designing a slicing algorithm according to the process of periodic and non-periodic layout in order that the processing pattern is automatically and quickly laid out on the surface of the complex component; by using the processing region division and pattern segmentation algorithm and setting the element type and combination in the processing pattern, obtaining the intersection of the element and the processing regionboundary and classifying the sub-pattern to the corresponding processing region; by using processing pattern real-time detection and online splicing manufacturing technology and by setting and installing a high-precision CCD camera, obtaining the video image of a processed part, recognizing the boundary of the processed pattern in real time based on an image processing algorithm, and then according to the boundary information of the pattern to be processed, predetermining the registration error of the two processing patterns, obtaining a required correction transformation parameter, and achieving online splicing manufacture in a laser processing process.

The invention discloses a self-adaptive image target area segmentation method based on SLIC, and relates to a superpixel segmentation technology. The objective of the invention is to solve the problemof low segmentation efficiency caused by the fact that parameters need to be manually input when super-pixel segmentation is carried out on a main body image containing a plurality of targets by a traditional super-pixel segmentation method. The method comprises the steps that firstly, super-pixel pre-segmentation processing is conducted on an image through SLIC, then super-pixel units are established with pre-segmented super-pixel points as center points, and the super-pixel measurement units comprise the gray level, the position and the Hash value; The measurement unit is used as a super-pixel parameter, and finally, through a distance-based clustering algorithm of adaptive parameters, the segmented too small regions are combined, so that the super-pixels are clustered into a determinedmain body and an obvious main body segmentation boundary. According to the method, a user does not need to carry out setting input, and the number of types of superpixels needing to be segmented is determined through a method of calculating the image complexity. The method is suitable for the fields of target recognition, mode recognition and artificial intelligence.

The invention discloses a thermal power plant time sequence variable abnormity monitoring method and system. The method comprises the steps of collecting time sequence data of thermal power plant analog signal monitoring variables; taking the whole time sequence data as an initial optimal PLR segment, calculating a decision coefficient of the optimal PLR segment, and taking the whole time sequencein the historical sample data as a first-layer original data segment; segmenting the original data segment to form two new data segments, and respectively calculating decision coefficients of the twonew data segments; judging whether the segments are allowed or not, and if yes, continuing to segment the segments; and judging whether the segmentation is finished or not to obtain the optimal PLR segmentation number. According to the method, the optimal segment number of the PLR is determined through a coefficient determination method, and the problems that an L method highly depends on a fitting error and left and right points in a time sequence are solved.

The invention discloses a road scene semantic segmentation method based on category grouping in abnormal weather. The method comprises the following steps: (1) preparing data; (2) road scene categories are grouped according to the importance of automatic driving safety in abnormal weather, it is guaranteed that important categories are separated from non-important categories, and a road scene semantic segmentation model is constructed according to the grouping result; (21) grouping the categories according to the importance of the automatic driving safety in the abnormal weather; (22) constructing a model according to a category grouping result; (3) inputting data into the model to obtain a segmentation result; (31) inputting data, and extracting data features; (32) acquiring full-category features; (33) encoding the category relationship by using the full-category features to obtain category relationship features; (34) completing category relationship feature decoding to obtain a segmentation result; (4) carrying out model iteration training; and (5) model testing. According to the method, the segmentation effect of important groups in abnormal weather can be ensured, and safe proceeding of an automatic driving task is ensured to the greatest extent. And meanwhile, an overall segmentation effect equivalent to that of a normal model can be obtained under the condition that the abnormal degree is relatively small.

The invention discloses a medical image segmentation model compression method, and belongs to the field of medical image processing. The method comprises the following steps: aiming at a medical image segmentation basic model, constructing a search space according to the number of convolution kernels used at each position in the model, aiming at a coding-decoding structure of a segmentation network, searching a sub-network with small calculation amount and high segmentation precision in the search space by using a symmetric neural network, and carrying out segmentation on the sub-network, wherein the coding and decoding structures are symmetrical, and weight sharing policies are used to mitigate computational costs and training resources when traversing the entire search space; finally, by using a knowledge distillation method in the network training process, the basic model serving as a teacher mode, the compressed sub-network serving as a student model, and completing the knowledge transfer between the basic model and the student model. Through neural network search and knowledge distillation, the calculation cost of network construction is greatly reduced on the premise of ensuring the segmentation effect of the medical image segmentation model, and the method can be applied to various medical image segmentation models.

The invention discloses a fruit segmentation method based on a sparse convolution kernel. According to the invention, the method comprises the steps: analyzing the distinction degree of a main objectin a fruit image under different color factors; selecting an appropriate color channel to reconstruct an image, and then proposing a sparse convolution kernel construction mode that elements are spaced from one another and are not adjacent to one another, and determining elements in a sparse convolution kernel by adopting a linear classifier; and finally performing convolution operation on the reconstructed image by adopting the sparse convolution kernel, so fruit segmentation is realized.

The invention discloses an activity signal separation method based on commercial Wi-Fi, and the method comprises the following steps: (10) CSI data collection: obtaining a group of CSI amplitude information Ak*t containing k subcarriers from a data packet in sampling time t; (20) active signal segmentation: calculating an average cross correlation coefficient of subcarriers in each second, and judging whether a person moves in the current environment or not, and if someone moves, continuing to calculate the amplitude difference of the interval time window within one second, and further refining the accurate moment of the user activity; (30) effective signal separation: iteratively calculating the minimum error of A-L-S by using an alternating direction multiplier method in an augmented Lagrange form, constraining the rank r of a background matrix L and the sparsity c of an activity matrix S, introducing bilateral random projection to reduce the calculated amount, performing wavelet transformation on the main components of the activity signals, and selecting specific frequency band activity signals generated by specific limb actions. The activity signal separation method is high inrobustness and high in operation efficiency.

The invention provides a cotton detection, segmentation and counting method and system, and belongs to the technical field of computer vision, and the method comprises the steps: obtaining a mask pattern matrix of an image background and a foreground, initializing a grabCut algorithm, and carrying out the segmentation; performing morphological processing, and extracting related attributes of each connected domain; combining the two connected domains meeting the condition into one region; for the connected domain whose area is larger than a preset first area threshold value and the ratio of the long axis to the short axis is larger than a preset threshold value of the ratio of the long axis to the short axis, dividing the connected domain into two independent areas; and performing cotton counting on the split single connected domain. According to the method, the accuracy is ensured while the model is kept light in volume, rapid and convenient to deploy; the segmentation effect of the algorithm is improved, and the efficiency of the subsequent merging and splitting processing process is improved; repeated operation is avoided, and the merging speed is increased; length detection does not need to be carried out on positions such as a bottleneck in the splitting process, it is guaranteed that the detected position with the highest brightness belongs to the two areas, and robustness is good.

The invention relates to the field of sludge treatment, in particular to a sludge treatment device which comprises a treatment chamber, and a pre-pressing box is fixedly connected into the treatment chamber. A pre-pressing device is arranged in the pre-pressing box body, and the pre-pressing device comprises two grinding rollers rotationally connected with the pre-pressing box body. A first conveying belt is arranged on a machine frame, a cutting device is arranged above the first conveying belt, and a gap is reserved between the cutting device and the first conveying belt. When sludge is recycled, the sludge can enter the pre-pressing box body from a feeding opening, and the pre-pressing device in the pre-pressing box body enables large sludge to be rolled to be in a thin state. The sludge falls onto the first conveying belt, and the first conveying belt conveys the sludge. The cutting device can play a role in cutting the sludge, and as the gap is reserved between the cutting device and the first conveying belt, the sludge cannot adhere to the cutting device when the sludge is cut, so that the cutting effect and the cutting efficiency are ensured.