Forage dominance recognition method based on convolutional neural network

Abstract

The invention relates to the technical field of image processing, and discloses a pasture dominance recognition method based on a convolutional neural network, which comprises the following steps: S1, dividing a monitored grassland into a plurality of grassland quadrats, photographing the grassland quadrats from an overlook angle to obtain at least two colorful quadrat images, performing image preprocessing on each quadrat image to obtain a quadrat image set; s2, constructing a pasture recognition neural network; and S3, inputting each quadrat image in the quadrat image set into a pasture recognition neural network to obtain a mask image indicating target pasture classification information, and calculating the coverage degree and dominance degree of each target pasture in each mask image according to the number of pixel points representing each target pasture type in each mask image. According to the method, the forage grass recognition precision can be improved, and reliable reference data is provided for grassland degradation monitoring.

Description

technical field [0001] The present invention relates to the technical field of image processing, in particular to a method for identifying pasture dominance based on a convolutional neural network. Background technique [0002] As one of the important components of the global ecosystem, grassland plays a key role in global climate change and ecological balance, but it is often threatened by grassland degradation due to factors such as overgrazing, mining, and human reclamation. Grassland degradation leads to miniaturization of plant communities and deterioration of soil properties, thus changing the composition of grassland plant community structure, which not only directly affects grassland livestock carrying capacity and grassland ecosystem productivity, but also causes frequent natural disasters such as land degradation, desertification, and sandstorms. Seriously affect environmental quality and ecological balance. Therefore, it is necessary to establish systematic monit...

AI Technical Summary

Problems solved by technology

Traditional monitoring methods of pasture population characteristics include on-the-spot visual inspection and direct harvesting methods. These traditional methods are gradually difficult to meet the current grassland survey work due to time-consuming and labor-intensive defects, and are affected by human factors, and their accuracy is difficult to obtain reliable guarantees.

Thanks to the development of multispectral remote sensing technology, through the multispectral remote sensing image system carried by satellites or UAVs, large-scale dynamic monitoring of grasslands can be realized, especially for macroscopic data such as vegetation coverage and biomass of grasslands. , non-destructive monitoring, eliminating the trouble of manual statistics. However, limited by the spatial resolution of remote sensing technology, coupled with the low overall height and mixed growth of forage populations, remote sensing technology is very useful in the classification of forage and the identification of the dominance of different types of forage. poor performance

Images