This study proposes PodNet, a lightweight deep convolutional network for accurate and fast soybean pod counting and localization from high-resolution images. PodNet employs a lightweight encoder and an efficient decoder to decode both shallow and deep information, reducing indirect interactions caused by information loss and degradation between non-adjacent levels. The model achieves an R² of 0.95 for soybean pod quantity prediction with only 2.48M parameters, significantly lower than the current state-of-the-art model YOLO POD. PodNet also has a much higher FPS, making it suitable for real-time applications. The model was evaluated on publicly available soybean pod detection and counting datasets, demonstrating superior performance in terms of accuracy, efficiency, and robustness. PodNet outperforms existing methods in various metrics, including precision, recall, mAP@0.5, and mAP@0.5:0.95. It also shows strong performance in complex environments with occluded pods. The model's lightweight architecture and high FPS make it suitable for deployment on low-cost devices. The study highlights the importance of efficient and accurate soybean pod counting for modern high-throughput plant phenotyping systems. Future research should focus on improving the model's performance in extreme conditions and exploring more sophisticated occlusion handling strategies.This study proposes PodNet, a lightweight deep convolutional network for accurate and fast soybean pod counting and localization from high-resolution images. PodNet employs a lightweight encoder and an efficient decoder to decode both shallow and deep information, reducing indirect interactions caused by information loss and degradation between non-adjacent levels. The model achieves an R² of 0.95 for soybean pod quantity prediction with only 2.48M parameters, significantly lower than the current state-of-the-art model YOLO POD. PodNet also has a much higher FPS, making it suitable for real-time applications. The model was evaluated on publicly available soybean pod detection and counting datasets, demonstrating superior performance in terms of accuracy, efficiency, and robustness. PodNet outperforms existing methods in various metrics, including precision, recall, mAP@0.5, and mAP@0.5:0.95. It also shows strong performance in complex environments with occluded pods. The model's lightweight architecture and high FPS make it suitable for deployment on low-cost devices. The study highlights the importance of efficient and accurate soybean pod counting for modern high-throughput plant phenotyping systems. Future research should focus on improving the model's performance in extreme conditions and exploring more sophisticated occlusion handling strategies.