Lightweight environment map prediction method and autonomous exploration system

Abstract

The present application relates to a kind of lightweight environment map prediction method and autonomous exploration system, comprising: obtaining confidence grid and corresponding true value grid and carrying out standardization and pretreatment, extract boundary point and cluster center, construct data set;Lightweight environment map prediction model is constructed based on UNet network, model adopts coding-neck-decoding structure, combined with residual connection, light convolution block and attention mechanism;Multi-objective loss function of adaptive weight mechanism is constructed, and environment map prediction model is optimized;The environment map prediction model after optimization is trained, and the local prediction subgraph of the environment map prediction model after training is obtained, and the prediction subgraph corresponding coordinate is projected into global coordinate and is fused to generate final prediction map.Compared with prior art, the present application has the advantages of high efficiency, high prediction accuracy and strong adaptability to complex environment of prediction model.

Images