The rapid development of Internet of Things (IoT), artificial intelligence (AI), edge computing, remote sensing and cloud platforms is reshaping agriculture from periodic observation toward continuous, data-driven management. This study surveys the convergence of IoT and deep learning (DL) for smart and sustainable agriculture, focusing on crop monitoring, disease and stress detection, irrigation, yield prediction, soil intelligence, resource optimization and environmental sustainability. The review finds that IoT provides fine-grained sensing and connectivity, while DL learns nonlinear spatial and temporal relationships from sensor streams, images, weather observations and historical farm records. CNNs are widely used for visual crop diagnostics, whereas LSTM/GRU, CNN-LSTM and Transformer models address temporal forecasting. Major gaps include fragmented single-task systems, laboratory-style datasets, limited multimodal integration, weak field-scale validation and inconsistent measurement of water, energy, chemical inputs and productivity. The proposed research direction is an integrated multimodal IoT-DL framework with edge-cloud intelligence, uncertainty-aware prediction and adaptive, sustainability-aware actuation.