This project explores the potential of machine learning and geospatial data in environmental monitoring for sustainable agriculture and ecological conservation. It focuses on the development of a system for crop detection and classification, providing insights for optimizing agricultural practices. Beyond its applications in agriculture, this technology holds significant potential for broader environmental monitoring initiatives. By correlating extracted features from satellite imagery with environmental parameters such as land cover, vegetation indices, and water quality, it can support habitat loss prevention, assess ecological changes, and contribute to overall ecological conservation efforts.
Cloud masks and cloud probabilities were then generated using the s2cloudless package (includedf in Eo Learn) to identify and filter out cloudy pixels, see Figure 5. By combining the satellite data with the known crop data, a model was trained to detect crops based on the extracted features. One of the features used to classify crops is the NVDI which can also be used to show vegetation density and health, see Figure 6.
The model utilized machine learning techniques to learn patterns and relationships between the satellite data and the crop types. The use of satellite data and the integration of known crop information in this approach allowed for data-driven analysis of crop distribution. This approach showcases the potential of combining remote sensing data with existing crop information to enhance crop monitoring and management practices.
The results of the crop detection model were generally positive, accurately identifying and labeling most crop sections, see Figure 7. The model demonstrated a strong capability to analyze visual patterns and features associated with various crops, leading to a successful recognition and classification of the majority of crops. However, there were instances where misclassifications occurred. The limited availability of specific crop data, despite using data from the state of California, contributed to these inaccuracies. More detailed and comprehensive crop data would have improved the model's ability to distinguish between visually similar crops. Incorporating detailed data would enhance the model's precision by capturing crop-specific characteristics.