2 results about "Logarithmic scale" patented technology

This invention relates to the field of UAV image processing technology, specifically to a method and system for diagnosing secondary degradation of alpine grasslands based on multi-source data. Existing technologies suffer from the following drawbacks: low image processing accuracy, large data processing errors, and a lack of spatiotemporal comparability. To address these shortcomings, this invention first acquires images using the hypotenuse target method and calculates the modulation transfer function characteristic parameters to construct a multi-scale Gaussian space. An adaptive threshold algorithm is used to determine and fix thresholds, achieving binarization of the multi-scale images. Connected components meeting certain conditions are extracted from the binary images at each scale, and their directional robust perimeters are calculated, forming a perimeter sequence. Based on the perimeter sequence, a logarithmic scale-perimeter curve is established, and the second derivative is calculated to determine the collapse scale and standard scale, calculating the geometric features of the connected components. Combining degradation judgment thresholds and proportional constraints, crack-type early degradation patches satisfying the condition with the maximum negative second-order curvature are identified, achieving automatic identification of secondary grassland degradation.

The application discloses a kind of based on double-flow self-adapting physical information fusion forging press energy consumption prediction system and method, comprising: based on physical principle to establish energy consumption mechanism model, and according to real-time state parameter calculation each component theoretical energy consumption value as mechanism data, mechanism data and real-time state parameter are selected for feature, respectively construct mechanism dataset and acquisition dataset, mechanism flow uses time series encoder based on mechanism dataset to carry out prediction, and introduce adaptive correction mechanism to compensate non-modeling energy consumption, while using logarithmic scale soft constraint to keep reasonable deviation of predicted value and theoretical value;Data flow uses time series encoder based on acquisition dataset to carry out prediction driven by pure data, and is trained by minimizing its prediction error, based on Monte Carlo Dropout quantification double-flow prediction uncertainty, generate dynamic fusion weight, and under the constraint of anti-degeneration regularization, the weighted fusion of two-flow predicted values is carried out, and the final comprehensive energy consumption predicted value is output.