6 results about "Local scale" patented technology

The application provides a packaging box anomaly detection method and device based on semantic constraints and geometric priors, and relates to the technical field of defect detection. The method extracts local geometric description information from three-dimensional point cloud data, inputs the information into a geometric perception measurement module to generate a geometric reliability coefficient, and uses the coefficient to adaptively modulate a local scale parameter predicted based on a direction vector to construct a multi-modal anomaly measurement constrained by a geometric prior. The method inputs a sample to be detected into a trained anomaly detection model, calculates local anomaly measurement values of each region through the multi-modal anomaly measurement, generates an anomaly score map reflecting the degree of anomaly, and performs threshold segmentation and connected component analysis to realize detection and positioning of packaging box anomalies. The method can realize high-precision packaging box anomaly detection without relying on a large amount of manually labeled data, and significantly improves the reliability and efficiency of packaging box quality detection.

The invention provides a packaging box anomaly detection method and device based on semantic constraint and geometric prior, and relates to the technical field of defect detection.The method comprises the steps that local geometric description information is extracted from three-dimensional point cloud data and input into a geometric perception measurement module to generate a geometric reliability coefficient; performing adaptive modulation on a local scale parameter based on direction vector prediction by using the coefficient, and constructing a multi-modal anomaly metric constrained by geometric prior; and inputting a to-be-detected sample into the anomaly detection model obtained by training, calculating a local anomaly metric value of each region through the multi-modal anomaly metric, generating an anomaly score graph reflecting an anomaly degree, and performing threshold segmentation and connected domain analysis to realize detection and positioning of the anomaly of the packaging box. According to the method, high-precision packaging box anomaly detection can be realized under the condition of not depending on a large amount of manual annotation data, and the reliability and efficiency of packaging box quality detection are remarkably improved.

The application discloses a caterpillar beam weld seam track fitting method based on an improved RANSAC algorithm and particularly relates to the technical field of weld seam track identification and automatic welding path planning. Standardized point cloud data is acquired. Welding spatter points are removed by combining the normal change rate of points with the extraction of weld protruding feature points through local height gradient analysis, a multi-scale point set structure of an overall scale point set and a local scale point set is constructed, a random consistency sampling is performed on the overall scale point set to fit a weld overall trend model, and a local track model set is generated on the local scale point set with the inner point set as a constraint. A track connection graph structure is constructed according to the local curvature change value and the direction continuity score, the screening and splicing of the local track model are completed, and a weld compensation track model is obtained. The compensation track model is subjected to continuous curve fitting to generate a weld center track curve. The application can realize high-precision weld seam track fitting in the presence of welding spatter and local deformation interference.

The application discloses a waveform-protected non-source wavelet local scale traveltime inversion method, relates to a non-source wavelet full waveform inversion method established by using a convolution and deconvolution method, and is combined with local scale traveltime inversion, and simultaneously solves the period jump of full waveform inversion and the source wavelet dependence problem. First, the influence of the source wavelet in the observation data and the simulation data is eliminated by using the convolution and deconvolution method, and the original seismic data waveform characteristics are protected. Secondly, local scale traveltime information of the observation data and the simulation data is extracted, and a waveform-protected non-source wavelet local scale traveltime inversion objective function is constructed. Then, the partial derivative of the waveform-protected non-source wavelet local scale traveltime inversion objective function to the velocity parameter is deduced in detail, and the corresponding underground velocity parameter gradient operator of the adjoint state method is deduced. Finally, the stable update of the velocity model parameter is realized by using an L-BFGS optimization algorithm. Finally, the effectiveness of the method is verified by the seismic data test.

The application provides a text pedestrian representation learning and matching method and system, comprising: 1) using a pre-trained ResNet model to generate a primary feature map for each input picture, and performing cutting processing on the feature map based on different scales; 2) using a Bottleneck Transformer as a representation learning network to perform self-attention calculation on different visual regions; 3) learning each word embedding through a pre-trained BERT model with fixed parameters; 4) further processing the word embedding through a hybrid branch network combining a residual network and a Transformer; 5) optimizing the representation of text and image by means of a cross-modal projection matching function (CMPM), and aligning the visual representation and the text representation from a local scale, a medium scale and a global scale respectively; 6) using the combined representation of the three scales as the final representation for retrieval. The application can extract good semantic information, has high accuracy and fast retrieval speed.

The application relates to the technical field of measurement remote sensing, and particularly discloses a cross-view geographical positioning method and system based on a multi-scale feature fusion pyramid, which comprises the following steps: acquiring target image data; extracting global scene features and multiple local detail features by adopting global and multiple local scale pooling on the target image data, performing convolution dimension reduction and fusion on the multiple local detail features, and obtaining local fusion features; performing linear transformation on the global scene features and the local fusion features to generate a feature mapping matrix, determining global attention weights and local attention weights based on the feature mapping matrix; determining spatial self-adaptive enhanced features based on the global attention weights and the local attention weights, performing fusion by adopting multi-feature residual connection, and obtaining a multi-dimensional feature vector; and outputting target positioning information based on the multi-dimensional feature vector and a preset satellite image database, so that the positioning accuracy is improved, and the matching process is optimized to reduce the complexity.