Remote sensing image culture pond detection method based on semantic segmentation

Abstract

The invention discloses a remote sensing image culture pond detection method based on semantic segmentation. The method is completed by four steps: firstly, cutting and partitioning a sub-meter levelremote sensing image, and storing index information of each image block at the same time; secondly, training a culture pond detection semantic segmentation model in advance until convergence to obtaina culture pond semantic segmentation model; inputting different remote sensing image blocks into the culture pond semantic segmentation model to obtain a remote sensing image block culture pond binary mask map; and finally, fusing the cut basic index information of the remote sensing image block, completing the combination of binary mask images, and generating a culture pond mask grid image. Theculture pond detection model is trained by using the deep learning technology, the extraction result is accurate, the robustness is good, and the adaptability is high; and a remote sensing image culture pond detection integrated process is designed, so that the artificial participation degree is reduced, and a certain practical value is achieved.

Description

technical field [0001] The invention relates to deep learning technology, mainly relates to pixel-level semantic segmentation, and is an end-to-end remote sensing image breeding pond detection method. 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 of operation, it can be divided into three categories: land, water and tidal flats. Land-based systems mainly include ponds, rice fields, and other land-based facilities; water-based development systems include bays, pens, cages, and raft cultures, usually located in coastal or inland waters with paddocks; Tidal flat-based culture systems include base pond culture and elevated pond culture. my country is the world's largest aquaculture country, and it is also the only country in the world where the aquaculture output exceeds the fishing output, and the scale of aquaculture is still growin...

AI Technical Summary

Problems solved by technology

[0004] The methods of extracting breeding ponds from remote sensing images mainly include visual interpretation, that is, based on interpretation signs (position, shape, size, tone, shadow, texture, etc.) There is a large workload, labor-intensive and time-consuming, and the timeliness of the quantitative analysis of massive spatial information is low; secondly, the method based on spatial structure analysis is not suitable for the extraction of breeding ponds with isolated distribution and small patches; in addition, object-oriented information The extraction method first divides the image into meaningful image objects, and then comprehensively uses information such as spectral characteristics, texture, shape, and proximity of ground objects to determine the category of the segmented object, which is greatly affected by feature selection and needs to be combined with visual interpretation. method

Images