VanillaNet-YOLOv8 Segment: Detection of Nano-Iron Oxide Regulation on Rice Seedling Growth Vitality under Salt Stress

Wei-Hang Jiang¹Hongyu Li¹Xuanhao Zhao¹Renhong Wang¹Xiuqing Fu^1*Zhibo Zhong²Ruxiao Bai²Yang Peng²Feng Pan³

• ¹College of Engineering, Faculty of Food and Engineering, Nanjing Agricultural University, Nanjing, China
• ²Institute of Farmland Water Conservancy and Soil-Fertilizer, Xinjiang, China
• ³Institute of Mechanical Equipment, Xinjiang Academy of Agricultural Reclamation Science, Xinjiang, China

Rice, a pivotal global food crop, faces a substantial threat from soil salinization during its growth cycle. The present study focuses on the regulatory effects of nano-iron oxide on the growth vitality of rice seedlings under salt stress, and constructs a technical system of "phenotype acquisition-model detection-vitality quantification". The present study utilised an independently developed high-throughput crop seedling phenotype detection system to obtain 3,888 full-time series growth images of rice seedlings over a period of 90 hours. The image quality was enhanced through preprocessing with a super-resolution algorithm (SSN). In order to address the challenges associated with detecting rice seedlings, which are characterised by their diminutive size and dense growth patterns, the YOLOv8-seg model has been enhanced. In this regard, the VanillaNet-YOLOv8 Segment model has been proposed. The VanillaNet concise backbone network was utilised to reduce computational complexity. The DualVit dual visual attention mechanism was introduced to decouple global semantics and local features to solve instance adhesion. A small object detection module was added to improve the recognition ability of weak seedlings, and the Real Value module was used to correct lens distortion and phototropic tilt to achieve accurate quantification of the true seedling length. The experimental findings demonstrate that the enhanced model attains a target detection accuracy (mAP50) of 98.4% and a segmentation accuracy (mAPmask50) of 96.4%, representing an improvement of 3.2% and 16.6%, respectively, over the original YOLOv8n-seg, while preserving its lightweight advantages. The "static vitality (average seedling length)-dynamic vitality (growth rate)" dual-index evaluation system was utilised to ascertain the most significant promoting effect on rice growth under salt stress (0-150 mmol/L NaCl). It was found that 300 mg/L nano-iron oxide had the most significant promoting effect on rice growth under salt stress, especially in terms of alleviating the inhibitory effect under severe salt stress. The present study provides an efficient and accurate technical framework for the evaluation of nanomaterial agricultural applications and the screening of salt-tolerant crops.