8results about How to "Improve image recognition" patented technology

The invention discloses a petroleum pipe microscopic fracture morphology identification method, system and device and a medium, and belongs to the technical field of failure analysis image process.The method comprises the steps that a deep learning classification network with a residual module is constructed and trained, and a deep learning classification network loaded with a network connection weight matrix is obtained; and finally, the deep learning classification network is utilized to efficiently and accurately identify the to-be-tested object region image. According to the method, massive data sets can be quickly processed and analyzed by using the high-speed processing capability of a modern computer, so that the efficiency is greatly improved. Compared with a traditional method, the automatic fracture classification method can reduce subjective interference of people and enhance reliability and consistency of results. And meanwhile, the identification method is simple, engineering application integration is facilitated, and good practicability and expansibility are achieved.

The invention discloses a stranded wire pitch measurement method based on a dynamic gray value under a linear array camera, and relates to the technical field of computer vision. Comprising the steps of S1, deploying a line-scan digital camera, an encoder and an encoding wheel according to the wire diameter and the production mode condition of a stranded wire, configuring parameters of the line-scan digital camera, adjusting the position of the line-scan digital camera, adjusting the encoder and the encoding wheel, S2, obtaining initial information of the line-scan digital camera and the stranded wire, S3, collecting images, S4, splicing the images, S5, processing the images, and S6, measuring the pitch, determining the pixel length of the stranded wire pitch in the image as a pixel distance; determining the actual length represented by each pixel according to the visual field of the line-scan digital camera and the resolution of the spliced image; and calculating the actual pitch: converting the proportion of the pixel distance to the actual length of each pixel to obtain the actual pitch of the stranded wire.

The invention relates to an optimization method for positioning measurement of an implanted crystalline lens after an ophthalmic operation, which comprises the following steps of: scanning an implanted astigmatism correction type artificial crystalline lens by using frequency sweep source optical coherence tomography (SS-OCT) to obtain image data containing a positioning mark, processing the image data by using a deep learning framework, and obtaining a positioning mark of the implanted astigmatism correction type artificial crystalline lens. Automatically identifying the astigmatism mark on the intraocular lens, and calculating the actual implantation axial position of the intraocular lens according to the identified astigmatism mark. The method has the advantages that axial position measurement can be completed without mydriasis, discomfort of a patient is reduced, dependence on operators is reduced, and measurement accuracy and efficiency are improved. The system comprises SS-OCT equipment and a deep learning processing module, and realizes automatic and non-invasive measurement of the implantation axial position of the artificial lens.

The present invention discloses a semantic soft label image recognition method and device with limited and unbalanced data, the method comprising: constructing a semantic soft label image recognition model; pre-training a self-supervised network on a large-scale text data set to obtain a word embedding module; using the word embedding module to generate a corresponding soft label for each category in the training data set; inputting the training data set into a feature extractor to obtain a feature vector, using the corresponding soft label to guide the training, and obtaining a trained semantic soft label image recognition model; inputting a test data set into the trained semantic soft label image recognition model for testing, and obtaining an image recognition result. The method trains a word embedding module on a large-scale text data set, and uses it to generate a corresponding soft label containing rich semantic information for each class of the data set, so as to help train an image recognition model with strong generalization performance under limited and unbalanced data conditions, thereby improving recognition performance.

This utility model discloses a limb fixation device for cervical spine imaging in rats, relating to cervical spondylosis research. It includes a fixation platform and a limiting component. The limiting component is installed in the middle of the top surface of the fixation platform, and a label is installed at one corner of the top surface. A supplementary light is located in the middle of one end of the top of the fixation platform, and fixing angle plates are symmetrically arranged on the left and right sides of the fixation platform. The limb fixation device for cervical spine imaging provided by this utility model has several adjustment holes around the placement area in the middle of the operating platform. Combined with the structural design of the limiting component, it can be appropriately adjusted within a certain range according to the rat's body structure, thereby ensuring structural fixation of the rat and guaranteeing the flexibility and convenience of the device. The label is used to help record data labels in the rat's images, improving image recognition.

The invention discloses a small target detection method, apparatus and device in a night environment, and a storage medium. The small target detection method in the night environment comprises the steps of obtaining a to-be-detected image; inputting the to-be-detected image into a trained detection model, performing convolution processing of a plurality of convolution layers, integrating extracted features through a full connection layer, and outputting an image detection result; wherein when a first convolution layer of the detection model performs convolution calculation, each local pixel block, covered by a convolution kernel, in a pixel region block of the to-be-detected image is adjusted according to the influence of a central pixel block of the pixel region block on other local pixel blocks, and the first convolution layer of the detection model performs convolution calculation; multiplying the convolution kernel to obtain a convolution result; and the central pixel block is set to be a fixed value 0 after being adjusted. By using the small target detection method in the night environment disclosed by the invention, the detection accuracy of the small target under the low light condition can be improved.

The invention relates to the technical field of X-ray imaging devices, in particular to a portable X-ray imaging matching device for quickly screening fish bone deformity, which comprises a fish body fixing and guiding module, a radiation source positioning module, a bone alignment imaging module, a gray scale fracture detection module and a deformity screening conclusion output module. According to the method, the influence of image interference information is reduced by comparing the position relation between the fitting point and the guiding track, pushing the tail of the fish body to be aligned with the guiding center line, continuously monitoring the offset position of the intersection point, correcting the corresponding relation between the irradiation path and the axis of the fish body, extracting the skeleton trunk area and removing edge fracture pixels; analyzing transverse contraction and included angle deviation trends in the image, delimiting abnormal position sections in the skeleton structure, highlighting the overall extension form and structural continuous trend of the skeleton contour in the image, and matching fish body individual differences and imaging environment changes to obtain a fish body image; and the image identification degree, the abnormal response speed and the result judgment stability degree in the malformation screening process are improved.

The application provides a target detection method based on differential feature compensation under polymeric multi-scale information, comprising constructing a multi-scale feature fusion module, selecting different receptive field convolution and various pooling methods to process input features to obtain differential features containing multi-scale information; constructing a parallel attention module to calculate spatial and channel attention maps in parallel, wherein various pooling methods are jointly used when inferring the attention map; combining the constructed multi-scale feature fusion module and parallel attention module in the best strategy to form a differential attention module for a feedforward convolutional neural network, and embedding the differential attention module into the network to obtain optimal performance. The application proposes a differential attention module (DAM) for a feedforward convolutional neural network, adopts a unique differential structure, and completes feature information compensation in the global through multi-scale feature fusion and attention mechanism to enhance the representation ability of features.