4results about How to "Heavy computation" patented technology

The present invention discloses a passenger pick-up and drop-off scheduling method for a customized bus system between urban areas, the steps of which are as follows: all stops are divided into a starting point area containing n starting points and a terminal area containing m terminal points, and a travel node sequence set Y={y 1 ,y 2 ,…,y n+m}; According to the inverse of the distance of each sequence, find the probability of each sequence being selected, sequence y i The probability of being selected is #imgabs0# and then based on the random number in the interval [0.0,1.1], the determination is to use the sequence y i nodes, or randomly add nodes to generate a new travel node sequence y 0 , and then iterate continuously until convergence; the present invention converts the path weight into a probability distribution, and on the basis of continuously passing on the advantageous subsequent nodes, adopts a 10% mutation probability for each node, which can quickly iterate and evolve; since the data exchange with the memory is reduced, the running speed can be improved; the iterative evolution of the present invention has strong cohesion and can conveniently realize parallel computing.

This invention discloses a light projector and a sensing system used in conjunction with the light projector. The light projector includes a light source and a diffraction optical element. The light source has a first light-emitting area and a second light-emitting area, and the diffraction optical element is disposed on the light source. The diffraction optical element has a first diffraction structure and a second diffraction structure, which overlap the first light-emitting area and the second light-emitting area, respectively. The first diffraction structure is different from the second diffraction structure.

The application discloses a satellite orbit ECI anchoring rendering method based on osgEarth and relates to the technical field of three-dimensional geographic information system visualization.The method comprises the following steps: storing orbit geometry in the ECI coordinate system statically, performing GMST rotation rendering through a MatrixTransform node every frame;monitoring the satellite position deviation, correcting the error through a double-path mechanism of anchor point translation and asynchronous resampling; supporting double-mode orbit display, coordinating the main thread and the background thread by using an asynchronous request identification matching filter mechanism; realizing earth occlusion culling by judging the intersection between the line of sight and the earth sphere in the GPU pixel shader; and driving the satellite propagation and the ground entity kinematics model by using a unified simulation time axis.The application solves the problems of orbit line drift, high-orbit penetration artifacts and time-space reference inconsistency, and realizes efficient, accurate and drift-free rendering and sky-ground collaborative display in a large-scale satellite scene.

The present invention discloses an adaptive multi-shape convolution method for target detection, the purpose of which is to solve the shortcomings of traditional convolution kernels that cannot change shape, cannot match target geometric deformation, have a single receptive field in the same network layer, and extract invalid information features other than the target. The method of the present invention mainly learns the kernel mask first, then uses the kernel mask to perform adaptive multi-shape convolution, and finally uses gradient back propagation to update the kernel mask. The adaptive multi-shape convolution method of the present invention enables convolution to adapt to changes in the geometric shape of the target, enhances the semantic extraction ability and feature expression ability of the convolution network, and only adds a small amount of parameters and computational complexity compared to standard convolution. It can be added to the backbone network of any target detection model, and only needs to replace the standard convolutions such as 3x3 and 5x5 in its network, which can greatly improve the average detection accuracy of network model target detection.