Research On Grassland Degradation Recognition Based On Image

Since modern times,with the continuous development of human society,the indiscriminate abuse of grassland resources and the impact of climate change,grassland degradation has become one of the increasingly prominent environmental problems,which has seriously affected human living environment,animal husbandry production,etc.And this situation is still getting worse.Therefore,it is imperative to protect the grassland ecology.Compared with traditional artificial field monitoring,modern rapid development of remote sensing technology enables researchers to obtain more comprehensive and abundant grassland data,combined with the latest computer science technology to establish a more efficient automatic grassland degradation monitoring system.In the beginning,the monitoring of grassland degradation was obtained by remote sensing technology to obtain vegetation sequence data,and the time series data of vegetation was analyzed and relevant indicators were calculated to reflect and evaluate the degree of regional grassland degradation.However,this method is greatly interfered by external factors,such as precipitation and temperature fluctuations,all of which affect the accurate judgment of grassland degradation.On the other hand,the lack of uniform and explicit evaluation indicators and degradation classification standards have also become obstacles to the evaluation of grassland degradation by vegetation sequence data.Therefore,if the high-resolution aerial image or satellite image can be used to automatically identify the degradation of grassland and image,such as region,area,and degree,it is of great significance to protect grassland ecology.In this paper,the method of deep learning is used to model the automatic recognition and segmentation of grass degraded images.The specific research contents are as follows:(1)Organize and analyze the existing samples of degraded grassland aerial photographs,and pre-process the data enhancement methods for the problems of unbalanced categories and too small sample numbers in the sample data set.The data set is 6:3:1.The scale divides the training set,the validation set,the test set,and normalizes the overall data before entering the network.(2)Research based on the full convolutional neural network to automatically identify and segment the degraded grassland image.The full convolutional neuralnetwork is a deep learning network model widely used in image semantic segmentation in recent years.Compared with traditional image processing methods and convolutional neural networks for image segmentation,the full convolutional neural network has end-to-end and arbitrary input and output results.A range of advantages such as input size and more accurate results.In this paper,we compare the common neural network semantic segmentation models and propose an improved model method,and compare it from multiple dimensions such as speed,accuracy,over-fitting,etc.In addition,this paper also explores the input size of different samples.The impact and selected the most appropriate sample input size for the current environment.(3)The paper further analyzes and optimizes the results.Firstly,the neural network visualization method is used to visually explore the proposed network model.The reliability of the neural network is proved to some extent by maximizing activation and interest highlighting;and the black edges and edge details existing in the actual test are Problems that are not smooth enough find the cause and are improved at the logical level of implementation.In this paper,deep learning is applied to image segmentation,and image recognition based on image is studied.Compared with traditional methods,it is more direct,reliable and has good application prospects.In this paper,the semantic segmentation model is further improved,and the samples and prediction methods are improved,so that it has better adaptability to the task of degraded grassland segmentation.