:: Volume 11, Issue 1 (9-2021) ::
JGST 2021, 11(1): 129-142 Back to browse issues page
Deep Learning-based Classification Method for Crop Mapping Using Time Series Satellite Images
N. Pourhasan, R. Shah-Hosseini *, S. T. Seydi
Abstract:   (1106 Views)
Awareness and knowledge about the cultivation pattern and the area under cultivation play an important role in agricultural land management and estimating net production. Combining the results of ground observations and measurements with remote sensing data can provide timely maps of crops. This is valuable for defining management units and achieving accurate information needed by farmers and planners. Most of the methods used to separate agricultural products do not work well when the cultivation pattern of different crops, such as wheat and barley, is very similar. Therefore, the purpose of this paper is to provide a deep learning-based classification method on satellite time-series images to produce a map of the exact area under the cultivation of different types of agricultural products with high technological similarity. For this purpose, the Landsat8 satellite time-series images were selected based on the region's crop calendar. Using a normalized vegetative differential indication index (NDVI) as a time series and a set of training data from various agricultural farms, the Convolutional Neural Network (CNN) was used to automatically generate a product map in the Chenaran region of North Khorasan Province. In order to separate the agricultural products in this study, a combination of supervised classification and visual correction has been used. In order to estimate the accuracy of the results, the maps produced with the ground control points were examined and the Kapa coefficient and overall accuracy were calculated. The results showed that the use of satellite series time data is highly effective in identifying and distinguishing different types of agricultural products. Also, the classification method based on the convolutional neural network with an overall accuracy of 95.76 has higher accuracy than other conventional methods such as random forest methods (overall accuracy: 89.85), backup vector machine (overall accuracy: 88.78), network Perspective neurons (overall accuracy: 85.75) and K have the closest neighbor (overall accuracy: 89.60) in separating and identifying agricultural products.
The present study was conducted to investigate the performance of the CNN classification method in producing the map of the area under cultivation of agricultural products in Chenaran city using Landsat8 satellite time-series images. The results showed that the use of the proposed CNN-based classification algorithm in identifying agricultural products according to the same training samples and two quantitative accuracy evaluation criteria showed better performance than other methods used and was able to effectively overcome to the existing challenges. It is very important to have accurate maps of the area under cultivation of various agricultural products in each region because this knowledge can be used in various fields such as improving the cultivation pattern of agricultural products, planning in the field of water resources required by the agricultural sector, Estimate the required budget to be used to allocate the machinery needed for each section. Therefore, using this method with optimal cost and time to produce a crop map is recommended. In future research, a combination of radar and optical images of Sentinel 1 and 2 satellites and deep learning networks will be used to achieve higher accuracy and better spatial resolution maps of the area under cultivation.
Keywords: Crop Mapping, Remote Sensing, Landsat8, Normalized Difference Vegetation Index (NDVI), Covolutional Neural Network (CNN)
Full-Text [PDF 1986 kb]   (650 Downloads)    
Type of Study: Research | Subject: Photo&RS

XML   Persian Abstract   Print

Rights and permissions
Creative Commons License This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.
Volume 11, Issue 1 (9-2021) Back to browse issues page