Early detection of fertilizer-induced stress in tomato plants is crucial for timely crop management interventions and yield optimization. Conventional optical methods detect fertilizer stress in young leaves with difficulty. This study proposes a novel, noninvasive technique for quantifying the density of trichomes-elongated hair-like structures found on plant surfaces-on young leaves using a smartphone. This method exhibits superior detection latency, enabling earlier and more accurate identification of fertilizer stress in tomato plants. Our approach combines augmented reality technology and image processing algorithms to analyze smartphone images of a specialized measurement paper. This measurement paper is applied to a tomato leaf to transfer trichomes onto its adhesive surface. The captured images are then processed through a pipeline involving region of interest extraction, perspective transformation, and illumination correction. Trichome detection and spatial distribution analysis of these preprocessed images yield a robust density metric. We validated our method through experiments on hydroponically grown tomatoes under varying fertilizer concentrations. Using leave-one-out cross-validation (LOOCV), our model achieves a mean area under the precision-recall curve of 0.824 and a receiver operating characteristic curve of 0.641 for predicting additional fertilization needs. Based on LOOCV, quantitative analysis revealed a strong relationship between trichome density and explanatory variables, including nitrate ion concentration, explaining 62.48% of the variation ($R^2 = 0.625$). The predicted and actual trichome densities were strongly correlated ($r = 0.794$). This straightforward and cost-effective method overcomes the limitations of traditional techniques, demonstrating the potential of using smartphones for practical plant nutrition diagnosis.

翻译：早期检测番茄植株的肥料诱导胁迫对于及时采取作物管理干预措施和优化产量至关重要。传统光学方法难以在幼叶中检测肥料胁迫。本研究提出了一种新颖的非侵入性技术，利用智能手机量化幼叶上毛状体（植物表面存在的细长毛发状结构）的密度。该方法展现出优异的检测延迟性，能够更早、更准确地识别番茄植株的肥料胁迫。我们的方法结合增强现实技术和图像处理算法，分析智能手机拍摄的专用测量纸图像。该测量纸被应用于番茄叶片，将毛状体转移到其粘性表面。捕获的图像随后通过一系列处理流程，包括感兴趣区域提取、透视变换和光照校正。对这些预处理图像进行毛状体检测和空间分布分析，得到一个稳健的密度指标。我们通过对不同肥料浓度下水培番茄的实验验证了我们的方法。使用留一法交叉验证（LOOCV），我们的模型在预测额外施肥需求方面，实现了精确率-召回率曲线下平均面积为0.824，接收者操作特征曲线下面积为0.641。基于LOOCV的定量分析揭示了毛状体密度与解释变量（包括硝酸根离子浓度）之间的强相关性，解释了62.48%的变异（$R^2 = 0.625$）。预测的与实际毛状体密度高度相关（$r = 0.794$）。这种简单且经济高效的方法克服了传统技术的局限性，展示了使用智能手机进行实际植物营养诊断的潜力。