Soil image classification is an important task in agricultural and environmental applications. With the advent of deep learning, image-based soil classification has become a popular research topic. Transfer learning is a widely used approach in image classification, where a pre-trained model is fine-tuned on a new dataset. However, traditional transfer learning approaches only fine-tune the last layer of the pre-trained model, which may not be sufficient for achieving high performance on a new task. In this paper, we propose HybridTransferNet, a novel hybrid transfer learning approach for image-based soil classification. HybridTransferNet fine-tunes a small number of earlier layers in addition to the last layer of a pre-trained ResNet50 model, resulting in improved classification performance compared to traditional transfer learning approaches. We evaluate HybridTransferNet on a soil classification dataset and report multiple performance metrics, including accuracy, precision, recall, and F1 score. Our experimental results show that HybridTransferNet outperforms traditional transfer learning approaches and achieves state-of-the-art performance on the soil classification task.