Floating HNS target detection method through combining multispectral image and deep learning method

Abstract

The invention discloses a floating HNS target detection method through combining a multispectral image and a deep learning method. The method comprises the steps of database preparation stage, targetdetection model construction stage and model application detection stage. The database preparation stage comprises an HNS characteristic reflection waveband database, a multispectral image database and a classification optimization waveband database, wherein the target detection model construction stage comprises region detection image data set construction, image preprocessing and labeling, target region detection model training and target category detection model training. The model application detection stage comprises detection image acquisition and preprocessing, target area segmentation,target category detection and visual detection. By utilizing the characteristic waveband images, the method has the advantages of strong pertinence, high image acquisition efficiency, high detectionaccuracy and the like, and is used for emergency detection of leakage accidents of transport ships of various HNS.

Description

technical field [0001] The invention relates to a floating HNS target detection method, in particular to a floating HNS target detection method combining multispectral images and deep learning methods. Background technique [0002] HNS (Hazards and Noxious Substances) water leakage accidents have brought a huge threat to the ecological environment and public safety. Since the color characteristics of HNS floating on the water surface are usually not obvious, from the ordinary RGB image, the difference between it and the background of the water body and the difference between the classes are small, which greatly increases the difficulty of rapid and automatic detection of the target. [0003] Currently, sensitive and accurate analytical techniques such as chromatography, spectrophotometry, and electrochemical methods are widely used in the detection and research of HNS. Most of these methods require sophisticated instruments and tedious sampling process, which limits their a...

AI Technical Summary

Problems solved by technology

However, compared with oil spill detection, HNS (such as benzene, toluene, xylene, vegetable oil, etc.) are often highly transparent liquids with little color difference from water, and the difference between classes is smaller, so it is difficult to automatically determine the leakage area using automated imaging methods. And HNS classification based on ordinary images is more difficult

Images