Embedded terminal reservoir water level early warning method based on semantic segmentation

Abstract

The invention discloses an embedded terminal reservoir water level early warning method based on semantic segmentation, which comprises the following steps of acquiring data of a reservoir in environment scenes such as multiple time periods, multiple weathers, multiple elements and the like based on an installed video site, and performing pixel-level semantic segmentation annotation on reservoir image data by using an annotation tool; performing format conversion and other preprocessing operations on the reservoir image data set, and performing batch enhancement on pictures and labels of the reservoir image data set through horizontal mirroring, random cutting and other modes; adopting a weighted cross entropy loss function and a multi-scale pooling feature fused FCN8s_loss_multi model torealize reservoir water area pixel-level segmentation; and deploying the FCN8s_loss_multi algorithm model on an embedded terminal to realize early warning identification of the water level of the reservoir. According to the method, the FCN8s_loss_multi algorithm after the weighted cross entropy loss function and the multi-scale pooling feature are fused is adopted in the semantic segmentation algorithm, so that the accuracy is high.

Description

technical field [0001] The invention relates to the technical field of water level early warning, in particular to an embedded terminal reservoir water level early warning method based on semantic segmentation. Background technique [0002] In the water conservancy industry, the water level of the reservoir and its early warning level are the key information for the safe operation and management of the reservoir. It is not only an important indicator for flood prevention such as reservoir flooding, but also important information for dam safety defense. In traditional applications, the water level of the reservoir mainly depends on manual inspection for on-site water gauge interpretation. In the past 20 years, some reservoirs have begun to use water level sensors for water level telemetry, but due to financial constraints, this method is mainly used in large and medium-sized reservoirs. With the rapid development of science and technology, embedded terminal technologies such...

AI Technical Summary

Problems solved by technology

[0005] In view of this, in order to solve the problems of relatively weak practicability, low efficiency and high cost of the water level early warning method in the prior art, the present invention proposes an embedded terminal reservoir water level early warning method based on semantic segmentation

Images