5983 results about "Artificial intelligence" patented technology

The present disclosure provides a method for predicting a target log frequency distribution within a depth interval. The method includes extracting drilling cuttings from a reference well and capturing a plurality of photographs of the drilling cuttings, wherein each photograph corresponds to a specific depth interval of a plurality of depth intervals. The method includes obtaining a reference target log from the reference well covering the plurality of depth intervals and extracting a plurality of image features from each photograph. The method includes capturing property variability within each depth interval including converting the plurality of image features and the reference target log into normalized frequency distributions. The method includes predicting model property variability by training a prediction model using the normalized frequency distributions of the image features as input data and the normalized frequency distribution of the reference target log as output data.

本发明涉及一种用于训练机器学习算法的计算机实现的方法，其中所述机器学习算法用于运行多个设备，所述方法包括以下步骤：‑ 提供并初始化待训练的模型，其中所述模型包括编码器和分类器；‑ 由所述设备提供用于本地训练的本地训练数据，并由中央计算单元提供用于全局训练的全局训练数据；‑ 由所述设备执行所述编码器的本地训练；‑ 由所述中央计算单元聚合经训练的编码器；‑ 由所述中央计算单元对所述分类器进行全局训练；以及‑ 由所述中央计算单元提供经训练的机器学习算法。

This application discloses a rapid method and related equipment for target segmentation in maintenance videos, relating to the field of target detection technology. The method includes: using a preprocessing module to downsample and segment the data to be processed, respectively, to obtain global input data and uniformly overlapping candidate blocks; using a global prediction module to generate a global coarse prediction result of tunnel defects based on the global input data, and using the global prediction module to generate a block importance probability distribution for each candidate block; using a differentiable Top-K sampling module to select the top K most important target blocks from each candidate block without repetition based on the block importance probability distribution; using a local fine segmentation module to segment and identify tunnel defects in each target block, obtaining a local fine prediction result; and using an adaptive aggregation module to fuse the global coarse prediction result and the local fine prediction result to obtain a tunnel defect segmentation map. This application can achieve high-precision and high-efficiency segmentation of tunnel defects.

This application discloses a vehicle and its control method, device, and medium, relating to the field of vehicle technology. The method includes: acquiring vehicle state information, road information, and surrounding vehicle state information; inputting the vehicle state information, road information, and surrounding vehicle state information into a first scene recognition model and a second scene recognition model respectively, to output a first scene recognition result and a second scene recognition result; fusing the first scene recognition result and the second scene recognition result based on the vehicle state information to obtain the target scene category of the vehicle and the confidence level of the target scene category; determining a target energy-saving control strategy for the vehicle based on the target scene category and the confidence level, and controlling the vehicle based on the target energy-saving control strategy.