This study aims to classify wheat and barley crops (irrigated and rainfed) in Famenin County, Hamadan Province, using deep neural networks and remote sensing data, to evaluate the accuracy and efficiency of multi-source satellite data integration for crop discrimination in semi-arid regions of Iran. In this study, Sentinel-1 (VV and VH bands) and Sentinel-2 (B4, B8, B11, and B12 bands) satellite data were utilized. Additionally, climatic parameters including Land Surface Temperature (LST), Elevation (DEM), and the MSAVI vegetation index were incorporated as complementary inputs to the model. For model training, the French RPG dataset combined with Iranian ground truth data from Famenin County was used for three years (2021, 2022, and 2023). The models employed included Unet, Unet++, ResNet, and MobileNet with DiceLoss and TverskyLoss cost functions. The results demonstrated that the ResNet-encoded Unet++ model achieved the highest performance with an overall accuracy of 81.2% and a Kappa coefficient of 0.76. Incorporating climatic parameters increased the overall accuracy by approximately 5.4%. The MSAVI index showed better performance in discriminating rainfed farms compared to NDVI. Furthermore, Sentinel-1 radar data had the greatest impact on the final accuracy.The findings revealed that the application of deep neural networks combined with multi-source satellite data can provide acceptable accuracy in agricultural crop classification in semi-arid regions of Iran. The integration of optical and radar data along with climatic parameters plays a significant role in improving classification accuracy. This approach can serve as an efficient tool for monitoring and managing agricultural crops. The combined use of Sentinel-1 and Sentinel-2 data along with climatic parameters and the MSAVI index within a deep neural network framework presents an innovative method for agricultural crop classification in semi-arid regions.