A Remote Approach to Cashew Orchard Detection: Leveraging Active Learning with Satellite Imagery in Guinea-Bissau

This study presents the first openly accessible, countrywide cashew orchard map for Guinea-Bissau, West Africa, created using a fully remote and cost-effective approach. The authors developed a scalable method based on Sentinel-2 satellite imagery and Machine Learning (ML) techniques to automatically detect cashew orchards across the entire country, expanding on previous regional analyses.

The methodology involved processing Sentinel-2 imagery from March 2017 to April 2021 in Google Earth Engine. The authors extracted 360 features per pixel, categorized into spectral features (including vegetation indices like NDVI, NBR, EVI, and EVI2), temporal features (from the Continuous Change Detection and Classification algorithm), and spatial features (from Gray-Level Co-Occurrence Matrix statistics). Feature selection reduced this to 67 features using an above-average importance threshold, which did not result in any performance drop.

A key innovation was the application of Margin-based Active Learning (AL) techniques to optimize the training dataset. The authors compared a Random Sampling (RS) dataset of 4498 pixels with a Margin Sampling (MS) dataset of 1909 pixels. The MS approach proved more efficient, achieving steeper performance increases in early iterations and a higher peak performance. The final classifier, a binary Support Vector Machine (SVM) trained on the MS dataset, achieved a Cashew F1-score of 85.7% and a Balanced Accuracy of 92.2% before post-processing.

After applying a sieve filter with a minimum polygon size of 350 pixels, the final cashew map achieved a Cashew F1-score of 89.5% and a Balanced Accuracy of 94.0%. The map shows that 25.9% of Guinea-Bissau's area (902,149 hectares) is identified as cashew orchards, with the highest concentration in the northwest quadrant. The authors also created two datasets and the 2021 cashew map with 10m spatial resolution, which are openly accessible through GitHub. This work provides a foundation for assessing the cashew sector's state and implementing measures to control its expansion, with potential applications in other West African countries facing similar biodiversity threats.

Improvements to AI Systems:

1. Active Learning for Label Efficiency: Integrate Margin-based Active Learning (MS) into any supervised classification pipeline to reduce training data requirements by 58% (1909 vs. 4498 pixels) while achieving higher peak F1-scores. The improved AI system can automatically select the most informative samples for human labeling, cutting annotation costs and time for large-scale land cover mapping.

2. Feature Reduction Without Performance Loss: Adopt an above-average importance threshold for feature selection (reducing 360 to 67 features) to eliminate redundant spectral, temporal, and spatial features. The improved AI system can run faster and with lower memory usage, enabling real-time or near-real-time classification on edge devices or in resource-constrained environments.

3. Temporal Feature Engineering for Cropland Detection: Use temporal features from the Continuous Change Detection and Classification (CCDC) algorithm to capture phenological cycles. The improved AI system can distinguish between permanent orchards and seasonal crops, improving robustness in regions with strong seasonal vegetation dynamics.

4. Post-Processing with Sieve Filters: Apply a minimum polygon size threshold (350 pixels) to remove speckle noise from pixel-wise classifications. The improved AI system can produce cleaner, more spatially coherent maps, reducing false positives and improving object-level accuracy (F1 from 85.7% to 89.5%).

5. Scalable Cloud-Based Pipeline: Replicate the Google Earth Engine workflow for countrywide analysis without field surveys. The improved AI system can automatically process multi-year satellite imagery (2017–2021) for any region, generating high-resolution (10m) land use maps at national scale, adaptable to other crops or land cover types.

6. Transferable Model for Biodiversity Monitoring: Use the trained SVM and feature set as a pre-trained baseline for neighboring West African countries. The improved AI system can quickly map cashew expansion or similar monoculture threats, enabling early warning systems for deforestation and biodiversity loss.

What the Improved AI System Can Do:

• Automatically map cashew orchards (or analogous crops) across entire countries with >89% F1-score using only freely available Sentinel-2 data.

• Operate with minimal labeled data, making it feasible for regions with scarce ground truth.

• Deliver updated maps annually (e.g., 2021, 2022, etc.) to track agricultural expansion and its environmental impact.

• Provide a reusable, open-source framework for other crop types (e.g., oil palm, rubber) by swapping training labels and adjusting features.

Cashew production is a widespread economic activity in Guinea-Bissau, as well as other countries in West Africa. However, unregulated cashew production can be directly associated with increasing regionwide deforestation rates, biodiversity losses, and a fragile economic structure. There is no nationwide database for listing or georeferencing cashew orchards, so there is a clear need to remotely map their locations. In recent years, multiple methods for detecting orchards have been developed, though they have only been applied on a regional level. This work expands regional analyses to a nationwide scale. It develops a scalable and cost-effective remote approach, based on Sentinel-2 satellite imagery, using Machine Learning techniques to detect cashew orchards automatically. Margin-based Active Learning techniques were employed to develop an optimal training set in terms of the number of points and their informativeness, leading to a cashew map with 94.0% balanced accuracy obtained entirely off-site. We created two datasets and a 2021 cashew map with 10m spatial resolution that are openly accessible through GitHub. The results demonstrate the possibility of a broader cashew orchard mapping, creating a new stepping stone for this environmental application.