Remote sensing image culture pond detection method based on instance segmentation

Abstract

The invention discloses a remote sensing image culture pond detection method based on instance segmentation. The method comprises the steps that firstly, a sub-meter remote sensing image is cut and partitioned, index information of each image block is stored at the same time, and the index information comprises geographic information; then instance segmentation is carried out on the remote sensingimage through a Mask R-CNN model, and calculation is carried out through multiple distributed GPUs; and in subsequent operation of instance segmentation, contour information is calculated according to the binary mask map of each culture pond position to obtain a polygonal vector of each remote sensing image block, spatial analysis and combination are performed on the polygonal vectors of the adjacent image blocks according to index information of the image blocks, and finally a culture pond vector Shp file of the whole remote sensing image is output. According to the invention, the culture pond instance segmentation model is trained by using the deep learning technology, the extraction result is accurate, the robustness is good, and the adaptability is high; in addition, the remote sensing image culture pond instance segmentation and production integrated process is designed and achieved, the manual participation degree is low, and the calculation efficiency is high.

Description

Technical field [0001] The invention belongs to the fields of computer vision, deep learning, and remote sensing image processing, and mainly relates to instance segmentation algorithms and related spatial topology analysis algorithms. Background technique [0002] Aquaculture refers to the commercial raising of aquatic organisms (including fish, molluscs, crustaceans and aquatic plants). According to the nature of the base surface of the operation, it can be divided into three categories: land, water and tidal flats. Land-based systems mainly include ponds, rice fields, and other facilities built on land; water-based research and development systems include bays, fences, cages and raft-type aquaculture, usually located in coastal or inland waters with enclosures; The tidal flat-based aquaculture system includes base pond aquaculture and high-level pond aquaculture. my country is the world's largest aquaculture country and the only country in the world whose aquaculture output e...

AI Technical Summary

Problems solved by technology

[0006] Although the semantic segmentation algorithm can efficiently and quickly extract the information of the breeding ponds and obtain the binary mask map of the breeding ponds in the remote sensing images, it cannot obtain the specific boundary area of ​​each breeding pond, and the value of the information is limited. In order to accurately obtain the information of each breeding pond Boundary area, a method for detecting breeding ponds in remote sensing images based on instance segmentation is proposed

Images