A Floating HNS Object Detection Method Combining Multispectral Images and Deep Learning Methods

Abstract

The invention discloses a floating HNS target detection method combining multispectral images and deep learning methods. The method includes database preparation, target detection model construction and model application detection stages; the database preparation stage includes HNS characteristic reflection band database, multispectral image Database and classification optimization band library; target detection model construction phase includes area detection image data set construction, image preprocessing and labeling, target area detection model training and target category detection model training; model application detection phase includes detection image acquisition and preprocessing, Object region segmentation, object class detection, and visualization detection. The invention utilizes the characteristic band image, has the advantages of strong pertinence, high image acquisition efficiency, high detection accuracy, etc., and is used for emergency detection of leakage accidents occurring on various HNS transport ships.

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