研究目的
To check whether different water and nitrogen treatments and the water-nitrogen coupling effect of plants could be correctly differentiated via chlorophyll a fluorescence image using imaging analysis and artificial neural networks.
研究成果
The method using chlorophyll a fluorescence imaging and RBF neural network effectively classified plant water and nitrogen statuses, with high accuracy for water treatments (above 85%) and slightly lower for nitrogen. The water-nitrogen coupling effect was studied, showing an average recognition rate of 69.8%, with lower accuracy under extreme conditions (e.g., high nitrogen and low water). The approach has potential for non-destructive monitoring but requires further validation with more samples and environmental factors.
研究不足
The study was conducted in a controlled laboratory environment, limiting generalizability to field conditions. Sample size for coupling experiments was smaller, reducing accuracy. External factors like temperature, humidity, and other nutrients were not considered, which could affect results. The method may not be fully optimized for real-time, in-situ monitoring without further development.
1：Experimental Design and Method Selection:
The study used chlorophyll a fluorescence imaging and artificial neural networks (specifically RBF, SVM, and BP networks) for classification. Time-resolved images were captured using an EMCCD camera with LED excitation at 460 nm and a 690 nm filter.
2：Sample Selection and Data Sources:
Scheffera octophylla (Lour.) Harms plants were used, cultivated in pots with clean sand. Ninety plants were divided into nine treatments combining three water levels (15 mL, 50 mL, 80 mL per day) and three nitrogen levels (0.75 g, 1.32 g, 1.95 g N per plot).
3：75 g, 32 g, 95 g N per plot).
List of Experimental Equipment and Materials:
3. List of Experimental Equipment and Materials: Equipment included LED arrays for excitation at 460 nm, an EMCCD camera (iXon Ultra 897, Andor), a 690 nm interference filter (690FS10-50, Andover Co. Ltd. USA), and a computer. Materials included plant samples, pots, sand, and nitrogen source (NaNO3).
4：3).
Experimental Procedures and Operational Workflow:
4. Experimental Procedures and Operational Workflow: Leaves were dark-adapted for 15 minutes before measurement. Fluorescence images were captured at different frame rates (e.g., 200 fps for fast section, 2 fps for slow section). Characteristic parameters (e.g., FI, FP, FS, FM, TFM, Fave, and coefficients from quadratic fitting) were extracted from the fluorescence induction curves.
5：Data Analysis Methods:
Data were split into training and testing sets (4:1 ratio). RBF, SVM, and BP neural networks were used for classification, with RBF selected as the best method based on recognition accuracy. Statistical analysis included calculating recognition rates and confusion matrices.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
