Global Vision Media Focus
The company was featured on CCTV a total of 15 times, and its reach through Party media, central media, and local official media reached a total of 50 million people.
Pixel-level identification and extraction of poppies based on multi-temporal and multispectral remote sensing images
Research Background
Poppies are a major source of drugs such as opium and morphine. While illegal poppy cultivation is strictly prohibited by Chinese law, sporadic illegal planting still exists in some remote areas. Traditional manual surveys are inefficient and have limited coverage, making them unsuitable for large-scale monitoring. Satellite remote sensing technology offers advantages such as wide coverage and...MultiphaseIts advantages of high objectivity make it an effective means of monitoring poppy cultivation.
Research Objectives
This study utilizesMulti-temporal Sentinel-2 L2A multispectral images (17th phase)Combined with known poppy plotsVector dataA remote sensing identification model for poppies was constructed that comprehensively considers spectral features, vegetation index features, phenological time series features, and flowering period texture features, so as to achieve accurate identification and spatial distribution mapping of poppy planting areas in the study area.
Study area
The target planting base is a legal poppy cultivation area. The survey shows that in 2026, the poppy flowering period in this area is June, and the flower color is white.
technical route
This study uses multi-temporal Sentinel2-L2A satellite imagery as the primary data source, constructing a multi-dimensional remote sensing feature set from four dimensions: spectrum, vegetation index, phenological time series, and flowering period texture. Preprocessing steps including sample equalization, feature selection, and feature standardization were completed.Random Forest (RF), Gradient Boosting (Gradient BoostingThree machine learning methods: 1) Support Vector Machine (SVM)Poppy identification models were constructed separately, and the optimal model was evaluated using cross-validation. The pixel-level poppy identification and extraction across the entire domain and the post-processing optimization of the results were completed. Finally, the binary classification map of poppies, the identification probability distribution map and the spatial distribution vector data were output, realizing accurate remote sensing identification and spatial quantitative extraction of poppy plots.
Figure 1.Poppy Remote Sensing Recognition Model Technical Roadmap
Technical methods
Random forest is a machine learning algorithm based on ensemble learning.By constructing a large number of "decorrelated" decision trees and integrating their predictions, the accuracy and robustness of the overall model are improved. Essentially, it's an ensemble of multiple decision trees, each of which is...Sampling with replacementIt is trained on a subset of data and a subset of random features. It can be used for both classification and regression.High accuracy, anti-OverfittingAdvantages include robustness to outliers and noise, and the ability to handle large-scale, high-dimensional data.
Figure 2. Core idea of random forest
Research Results
Figure 3.Poppy remote sensing identification binary classification image
Figure 4.Poppy remote sensing identification probability distribution map
Analysis of advantages and limitations
Advantages Analysis
1. Multidimensional feature fusionThe four characteristics are: comprehensive spectrum, vegetation index, phenological time series, and texture.
2. In-depth utilization of the red-edge band: The key red-edge band combination (684/706/725nm) confirmed by literature.
3. Multi-temporal phenological modeling: Use the time sequence information of the complete growing season to distinguish poppies from other plants that may cause confusion.
Limitations Analysis
1. Limited training samplesOnly a portion of the poppy fields are known, and the model's generalization ability needs further verification.
2. Spatial resolution limitationsSentinel-2's 10m resolution has limited ability to identify extremely small patches.
3. Risk of confusion between categoriesWhite rapeseed flowers and other similar flowers may cause misidentification during their flowering period.