To address the problems of difficult identification and low detection accuracy of various nutrient deficiencies in actual lettuce cultivation environments, this paper proposes an improved YOLOv8n model for lettuce nutrient deficiency target detection, and names this algorithm YOLOv8n-SCD. To enhance detection accuracy and inference speed, the improvements include: (1) introducing the SimAM attention mechanism into the backbone network, which builds a parameter-free saliency enhancement module to effectively improve the model’s perception of local details; (2) integrating ConvNeXtV2 into the backbone network to construct the C2f-ConvNeXtV2 model structure, which enhances multi-scale perception and fine-grained semantic modeling capabilities while maintaining network efficiency; (3) By employing the Distance-IoU (DIoU) loss function, the precision of the localization component within the lettuce nutrient deficiency detection model can be significantly enhanced, particularly in scenarios where the gap between the centers of the actual and predicted bounding boxes is considerable. Through these improvements, YOLOv8n-SCD has increased the mAP50 by 2.2% in the lettuce nutrient deficiency detection task, achieving effective optimization of the original algorithm. Experimental results show that the YOLOv8n-SCD model has improved recognition accuracy, providing an effective solution for agricultural robots to detect nutrient deficiencies in lettuce.