Estimation of Iron Content in Apple Leaves Using an Artificial Neural Network and Image Processing Model

Document Type : Research Paper

Authors

• Hojjat Sadeghi ¹
• Ebrahim Sepehr ²
• Aydin imani ¹

¹ Dept. of Soil Science, Faculty of Agriculture, Urmia University, Urmia, Iran

² Department of Soil Science, Urmia University, Urmia, Iran

Abstract

Keywords

• Apple leaf
• Machine learning
• Active iron
• Image processing

Main Subjects

• ..

Introduction

Iron deficiency is an important nutritional disorder in fruit trees growing on calcareous soils, where a high pH limits the availability of iron. This leads to reduced growth, leaf chlorosis and yield loss. Conventional detection methods are accurate but time consuming and costly. Recent advances in digital imaging and machine learning offer a fast, cost-effective and non-destructive alternative. This study investigates the use of leaf color features and machine learning to estimate active iron content in apple leaves under field conditions.

Methods

A total of 1575 images of apple leaves were collected from about 50 orchards around Urmia, Iran. The samples represented four levels of iron deficiency: none, low, moderate, and severe. The images were captured using a Samsung smartphone camera mounted 15 cm above the leaves in a custom-built imaging chamber under uniform lighting. The images were saved in 24-bit RGB format with JPEG enhancement. The active iron content and total iron content of the corresponding leaf samples were measured by atomic absorption spectroscopy. Active iron was extracted using the 1.5% phenanthroline method (pH = 3), while total iron was measured by dry ashing followed by acid digestion. To extract color features, images were first preprocessed to remove background noise using the Excess Green (ExG) index. Subsequently, color features were extracted from the RGB, HSV, Lab, and NTSC spaces as well as eight vegetation indices (e.g., GMR, GDR, DGCI, NRI, NGI). The mean value of each color channel was used as input. The most important features were selected by univariate regression analysis. A multilayer perceptron (MLP) neural network was developed in MATLAB 2017 using the Levenberg–Marquardt training algorithm. The dataset was randomly split into 70% training, 15% validation, and 15% test. Model accuracy was assessed using R², root mean square error (RMSE), and mean absolute percentage error (MAPE).

Results

A univariate regression analysis was performed to evaluate the relationship between color features and active iron content. Among the RGB channels, R and G showed the highest correlation with active iron content (R² = 0.67). The H and V channels in HSV and the L and b channels in Lab also showed acceptable accuracy, indicating the importance of brightness and hue changes due to iron deficiency. Indices such as NRI and DGCI also showed promising performance. However, no significant correlation was found between total iron content and any color feature. Using the best performing features (R, H, Y, I), a multilayer perceptron neural network (MLP) with a 4–25–1 architecture was developed. The model was trained using the Levenberg–Marquardt algorithm in MATLAB. The prediction accuracy reached R² = 0.80 with RMSE = 1.56 and MAPE = 25.03% for the entire dataset. Histogram and scatter plots confirmed that the predicted values matched the actual values very well, especially at low iron levels. A comparison with multiple linear regression (R² = 0.74, RMSE = 1.33, MAPE = 31.54%) confirmed the superior performance of the neural network in modeling complex nonlinear relationships between leaf color and active iron content.

Conclusion

This study confirmed that active iron in apple leaves can be accurately estimated using image-based color features and an artificial neural network. The most important color channels (R, H, Y, I) showed the strongest correlations with the active iron, while the total iron showed no significant correlation. These results underline the potential of a low-cost, smartphone-based image analysis system for the early diagnosis of iron deficiency in orchards. This approach could replace costly laboratory testing and enable scalable, real-time nutrient monitoring to support precision agriculture. Future developments could focus on extending the model to other micronutrients and integrating the system into mobile applications for farmers and extension specialists to bridge the gap between advanced analytics and on-farm decision making.

Funding

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

Authorship contribution

Conceptualization, E.S. and A.I.; methodology, E.S., A.I. and H.S.; software, A.I.; validation, E.S. and A.I.; formal analysis, H.S. and A.I.; investigation, H.S. and A.I.; resources, E.S.; data curation, E.S. and H.S; writing—original draft preparation, H.S. and A.I.; writing—review and editing, E.S. and A.I.; visualization, H.S. and A.I.; supervision, E.S. and A.I.; project administration, E.S.; funding acquisition, E.S. All authors have read and agreed to the published version of the manuscript.

Declaration of Generative AI and AI-assisted technologies in the writing process

The authors declare that they did not use of generative AI and AI-assisted technologies to write the manuscript.

Data availability statement

Data available on request from the authors.

Acknowledgements

The authors sincerely appreciate the support and facilities provided by the Department of Soil Science Engineering, Faculty of Agriculture, Urmia University, Iran.

Conflict of interest

The authors declare no conflict of interest.