Sustainability of agriculture has remained as a pillar of food security in the world, economic growth, and management of the environment. The integration of intelligent agriculture with modern computational methods, namely deep learning (DL), Internet of Things (IoT), and advanced big-data analytics, is a new milestone in the exact management of crops in the modern world. Conventional agriculture and classical machine learning techniques usually do not represent the complex, dynamic and large-scale nature of agricultural data, especially under the real field conditions. Here, we review 25 recent studies on DL-based plant disease detection and crop classification in a systematic manner covering a variety of models that include convolutional neural networks (CNN’s), hybrid CNN-LSTM/BiLSTM neural architectures, Transformer models, YOLO-based real-time detection schemes and IoT-integrated systems. We identify some of the main trends including gradual shifts from CNN-only models to the hybrid and Transformer-based ones, real-time detection systems, IoT-integrated systems for continuous monitoring, and improved model robustness and generalization. In spite of these advances, there still persist several challenges including limited dataset diversity, high computational burden, lack of interpretability and diminished performance under variable field conditions. By consolidating methods, datasets, tools, crops of interest, and performance metrics, this review points out important research gaps and actionable insights for constructing scalable, interpretable and field deployable DL systems, which would further the development of smart agriculture and sustainable crop management.