Gürültülü Etiketler Altında Derin Öğrenme Tabanlı Ürün Beyanı Doğrulama: Parsel Düzeyinde Risk Önceliklendirme için Kayıp Fonksiyonları ve Mimarilerin Karşılaştırmalı Analizi

Abstract

Tarımsal destek politikalarının etkin yönetimi, çiftçi beyanlarının güvenilirliğine doğrudan bağlıdır. Türkiye'de Çiftçi Kayıt Sistemi (ÇKS) aracılığıyla yürütülen parsel bazlı ürün beyanları, saha düzeyinde sistematik doğrulamadan yoksundur; bu durum beyan verilerinin eğitim etiketi olarak kullanıldığı derin öğrenme modellerinde yapısal bir gürültü kaynağı oluşturmaktadır. Bu çalışmada, Çanakkale ili Biga ilçesindeki 15.000 tarım parseli için Sentinel-2 çok zamanlı verileri kullanılarak ÇKS beyanlarını doğrulamaya yönelik derin öğrenme tabanlı bir sınıflandırma ve risk değerlendirme sistemi geliştirilmiştir. Çalışmanın özgün katkısı iki eksende yoğunlaşmaktadır: (i) Symmetric Cross Entropy (SCE) kayıp fonksiyonunun uzaktan algılama tabanlı parsel sınıflandırmasına ilk kez uygulanması ve (ii) sınıflandırma çıktılarının parsel bazlı risk skoruna dönüştürülerek saha doğrulama önceliklendirme çerçevesine entegrasyonu. LSTM ve 1D CNN-LSTM mimarileri, standart Categorical Cross Entropy (CCE) ile SCE kullanılarak karşılaştırmalı olarak değerlendirilmiştir. On Sentinel-2 spektral bandı ile NDVI, EVI, NDRE ve LAI vejetasyon indekslerinden oluşan 14 özellikli haftalık kompozitler, 30 zaman adımı boyunca girdi olarak kullanılmıştır. En yüksek performans, 1D CNN-LSTM + SCE modelinden elde edilmiştir (Macro F1: 0.741; OA: %86.5). SCE, azınlık sınıflarında belirleyici iyileşmeler sağlamıştır: Domates precision değeri 0.16'dan 0.70'e, Biber precision değeri 0.41'den 0.76'ya yükselmiştir. Risk skorlama mekanizması, 15.000 parselin 1.738'ini (%11.6) yüksek riskli olarak işaretleyerek saha denetim kaynaklarını 8.6 kat daha verimli kullanmaya imkân tanımaktadır. Coğrafi kümelenme analizi, belirli köylerde uyumsuzluk oranlarının %48'e ulaştığını ortaya koymuştur. Bulgular, etiket gürültüsü içeren tarımsal sınıflandırma problemlerinde kayıp fonksiyonu seçiminin mimari seçiminden daha belirleyici olabileceğini göstermektedir.

Keywords

• Sentinel-2
• Tarımsal Uzaktan Algılama
• Etiket Gürültüsü
• Beyan Doğrulama
• Derin Öğrenme
• Risk Skorlama

Deep Learning-Based Crop Declaration Validation with Noisy Labels: A Comparative Study of Loss Functions and Architectures for Parcel-Level Risk Prioritization

Abstract

Accurate verification of farmer-declared crop types is essential for the reliable administration of agricultural subsidy programs. In Turkey, parcel-level crop declarations submitted through the Farmer Registration System (FRS/ÇKS) are not subject to systematic field-level verification, introducing structural label noise into deep learning models that rely on these declarations as training supervision. This study presents a deep learning-based crop classification and risk assessment framework applied to 15,000 agricultural parcels in the Biga district of Çanakkale province, Turkey, using Sentinel-2 multi-temporal imagery to validate FRS/ÇKS declarations. The study advances the state of the art along two axes: (i) the first application of Symmetric Cross Entropy (SCE) loss to remote sensing-based parcel classification, and (ii) the integration of classification outputs into a parcel-level risk scoring framework for field verification prioritization. LSTM and 1D CNN-LSTM architectures were comparatively evaluated under both standard Categorical Cross Entropy (CCE) and SCE loss functions. Weekly median composites comprising 14 features — ten Sentinel-2 spectral bands together with NDVI, EVI, NDRE, and LAI vegetation indices — were used as input across 30 temporal steps. The highest performance was achieved by the 1D CNN-LSTM + SCE configuration (Macro F1: 0.741; OA: 86.5%). SCE yielded decisive improvements for minority classes: tomato precision increased from 0.16 to 0.70, and pepper precision from 0.41 to 0.76. The risk scoring mechanism flagged 1,738 of 15,000 parcels (11.6%) as high-risk, enabling an 8.6× more efficient allocation of field inspection resources. Geographic clustering analysis revealed discrepancy rates reaching 48% in certain villages. These findings demonstrate that, in agricultural classification problems with structural label noise, loss function selection can be more determinative than architectural choice.

Keywords

• Sentinel-2
• Agricultural Remote Sensing
• Label Noise
• Declaration Validation
• Deep Learning
• Risk Scoring

References

1. Clevers, J. G. P. W., & Gitelson, A. A. (2013). Remote estimation of crop and grass chlorophyll and nitrogen content using red-edge bands on Sentinel-2 and -3. International Journal of Applied Earth Observation and Geoinformation, 23, 344–351.
2. Di Martino, T., Guinvarc'h, R., Thirion-Lefevre, L., & Colin, E. (2023). FARMSAR: Fixing agricultural mislabels using Sentinel-1 time series and autoencoders. Remote Sensing, 15(1), 35.
3. FAO. (2024). FAOSTAT statistical database. Food and Agriculture Organization of the United Nations. https://www.fao.org/faostat/
4. Ghosh, A., Kumar, H., & Sastry, P. S. (2017). Robust loss functions under label noise for deep neural networks. Proceedings of the AAAI Conference on Artificial Intelligence, 31(1), 1919–1925.
5. Ghosh, A., Manwani, N., & Sastry, P. S. (2015). Making risk minimization tolerant to label noise. Neurocomputing, 160, 93–107.
6. Gorelick, N., Hancher, M., Dixon, M., Ilyushchenko, S., Thau, D., & Moore, R. (2017). Google Earth Engine: Planetary-scale geospatial analysis for everyone. Remote Sensing of Environment, 202, 18–27.
7. Haboudane, D., Miller, J. R., Pattey, E., Zarco-Tejada, P. J., & Strachan, I. B. (2004). Hyperspectral vegetation indices and novel algorithms for predicting green LAI of crop canopies: Modeling and validation in the context of precision agriculture. Remote Sensing of Environment, 90(3), 337–352.
8. Hochreiter, S., & Schmidhuber, J. (1997). Long short-term memory. Neural Computation, 9(8), 1735–1780.

9. Huete, A., Didan, K., Miura, T., Rodriguez, E. P., Gao, X., & Ferreira, L. G. (2002). Overview of the radiometric and biophysical performance of the MODIS vegetation indices. Remote Sensing of Environment, 83(1–2), 195–213.
10. Immitzer, M., Vuolo, F., & Atzberger, C. (2016). First experience with Sentinel-2 data for crop and tree species classifications in Central Europe. Remote Sensing, 8(3), 166.
11. Interdonato, R., Ienco, D., Gaetano, R., & Ose, K. (2019). DuPLO: A dual view point deep learning architecture for time series classification. ISPRS Journal of Photogrammetry and Remote Sensing, 149, 91–104.
12. Ioffe, S., & Szegedy, C. (2015). Batch normalization: Accelerating deep network training by reducing internal covariate shift. Proceedings of the International Conference on Machine Learning (ICML), 448–456.
13. Kang, J., Fernandez-Beltran, R., Duan, P., Kang, X., & Plaza, A. J. (2021). Robust normalized softmax loss for deep metric learning-based characterization of remote sensing images with label noise. IEEE Transactions on Geoscience and Remote Sensing, 59(10), 8798–8811.
14. Kingma, D. P., & Ba, J. (2015). Adam: A method for stochastic optimization. Proceedings of the International Conference on Learning Representations (ICLR).
15. Pelletier, C., Valero, S., Inglada, J., Champion, N., & Marais Sicre, C. (2017). Effect of training class label noise on classification performances for land cover mapping with satellite image time series. Remote Sensing, 9(2), 173.
16. Pelletier, C., Webb, G. I., & Petitjean, F. (2019). Temporal convolutional neural network for the classification of satellite image time series. Remote Sensing, 11(5), 523.
17. Rouse, J. W., Haas, R. H., Schell, J. A., & Deering, D. W. (1974). Monitoring vegetation systems in the Great Plains with ERTS. Proceedings of the Third Earth Resources Technology Satellite-1 Symposium, 309–317.
18. Schmedtmann, J., & Campagnolo, M. L. (2015). Reliable crop identification with satellite imagery in the context of Common Agriculture Policy subsidy control. Remote Sensing, 7(7), 9325–9346.
19. Sitokonstantinou, V., Papoutsis, I., Kontoes, C., Lafarga Arnal, A., Armesto Andrés, A. P., & Garraza Zurbano, J. A. (2018). Scalable parcel-based crop identification scheme using Sentinel-2 data time-series for the monitoring of the Common Agricultural Policy. Remote Sensing, 10(6), 911.
20. Song, H., Kim, M., Park, D., Shin, Y., & Lee, J.-G. (2022). Learning from noisy labels with deep neural networks: A survey. IEEE Transactions on Neural Networks and Learning Systems, 34(11), 8135–8153.
21. Sonobe, R., Yamaya, Y., Tani, H., Wang, X., Kobayashi, N., & Mochizuki, K. (2018). Crop classification from Sentinel-2-derived vegetation indices using ensemble learning. Journal of Applied Remote Sensing, 12(2), 026019.
22. Şimşek, F. F., & Durduran, S. S. (2022). Land cover classification using Land Parcel Identification System (LPIS) data and open source Eo-Learn library. Geocarto International, 38(1), 2146760.
23. Tufail, R., Tassinari, P., & Torreggiani, D. (2025). Deep learning applications for crop mapping using multi-temporal Sentinel-2 data and red-edge vegetation indices: Integrating convolutional and recurrent neural networks. Remote Sensing, 17(18), 3207.
24. United Nations. (2015). Transforming our world: The 2030 agenda for sustainable development (A/RES/70/1).
25. Veloso, A., Mermoz, S., Bouvet, A., Le Toan, T., Planells, M., Dejoux, J. F., & Ceschia, E. (2017). Understanding the temporal behavior of crops using Sentinel-1 and Sentinel-2-like data for agricultural applications. Remote Sensing of Environment, 199, 415–426.
26. Vuolo, F., Neuwirth, M., Immitzer, M., Atzberger, C., & Ng, W. T. (2018). How much does multi-temporal Sentinel-2 data improve crop type classification? International Journal of Applied Earth Observation and Geoinformation, 72, 122–130.
27. Wang, Y., Ma, X., Chen, Z., Luo, Y., Yi, J., & Bailey, J. (2019). Symmetric cross entropy for robust learning with noisy labels. Proceedings of the IEEE/CVF International Conference on Computer Vision, 322–330.
28. Zhang, Z., & Sabuncu, M. R. (2018). Generalized cross entropy loss for training deep neural networks with noisy labels. Advances in Neural Information Processing Systems, 31, 8792–8802.
29. Zhong, L., Hu, L., & Zhou, H. (2019). Deep learning based multi-temporal crop classification. Remote Sensing of Environment, 221, 430–443.