Document Type : Original Article

Authors

• Amir Aghabalaei ¹
• Jahede Tekyeh Khah ²
• Shler Katorani ³
• Ali Mohtadi ⁴

¹ Imam Hossein Comprehensive University

² Jihad University of Kurdistan

³ Shahid Beheshti University

⁴ Hekmat Higher Educational Institution

Abstract

The aim of this study is to develop and evaluate a hybrid deep learning-based model for accurate classification of strawberry crops using Sentinel-2 satellite time-series imagery and indigenous phenological data in Kurdistan Province. The primary motivation for designing the model lies in the spatial identification of crops with limited distribution and specific growth cycles.
Ground truth data were collected and labeled during the 2022 growing season using GPS and the regional agricultural calendar. Selected Sentinel-2 images temporally aligned with the phenological stages of strawberry growth underwent geometric, radiometric, and atmospheric preprocessing. These data were then fed into a Convolutional Neural Network (CNN) consisting of three convolutional layers, one pooling layer, two fully connected layers, and a Softmax output function. To address class imbalance, class weighting techniques and grid search optimization were employed. Data augmentation was also utilized to enhance the model's generalizability. The training process was conducted using the Adam optimizer with a learning rate of 0.0001 over 150 epochs. The proposed model successfully classified 15 land use/land cover classes with an overall accuracy of 96.57% and a Kappa coefficient of 0.8582. For the target class—strawberry—the F1-score reached 86.4%, indicating a favorable balance between precision and recall. The model demonstrated strong performance in identifying crops with limited cultivated areas and irregular spatial structures, although accuracy declined in classes such as greenhouses and rangelands due to high spectral heterogeneity. Error analysis revealed that most misclassifications occurred at spectrally similar boundaries and during climatic fluctuations in the growth period.

Keywords

• Crop classification
• Strawberry
• Convolutional Neural Network
• Time series
• Deep learning

Main Subjects

• Remote sensing and Gis