نوع مقاله : مقاله پژوهشی
نویسندگان
1 گروه علوم خاک، دانشکده کشاورزی، دانشگاه ارومیه، ارومیه، ایران
2 گروه پژوهشی شیمی تجزیه، جهاد دانشگاهی آذربایجان غربی، ارومیه، ایران
چکیده
کلیدواژهها
موضوعات
عنوان مقاله [English]
نویسندگان [English]
Accurately and promptly monitoring the nutritional conditions of fruit orchards is crucial for providing optimal fertilizer recommendations, which in turn improves yield and enhances the quality of agricultural products. The current laboratory methods used to evaluate nutritional condition in fruit trees are expensive, challenging, time-consuming, and require an expert. In this study, image processing methods and neural network models was utilized to determine the stages of iron deficiency in peach trees. Therefore, a database containing 800 images of peach leaf samples was acquired. These images were then classified into four categories using the KNN clustering method: no deficiency, low deficiency, moderate deficiency, and severe deficiency. The preprocessing, feature extraction, and modeling operations were performed in the MATLAB software, version 2017. Features such as mean and standard deviation were extracted from the RGB, HSV, and Lab color space components of each image. Subsequently, the principal component analysis (PCA) algorithm was applied to the feature vector. To determine the optimal structure of the network, criteria including precision, accuracy, recall, and the F1-score were evaluated. These criteria helped ascertain the number of optimal inputs and the corresponding number of neurons for each combination of input features (PCs). Results indicated that the neural network model, structured as 6-36-4, achieved an accuracy of 89.73 ± 0.54%, precision of 89.59 ± 0.57%, recall of 89.52 ± 0.51%, and an F1-score of 89.55 ± 0.54% in detecting levels of iron deficiency in peach tree leaves. The findings from the confusion matrix and the developed model reveal that this method can effectively and efficiently detect the severity of iron deficiency in peach tree leaves.
کلیدواژهها [English]
Detection of iron deficiency in peaches using image processing and artificial neural network model
EXTENDED ABSTRACT
Among micronutrients, iron deficiency is regarded as a significant nutritional disorder in orchards established on calcareous soils. Employing destructive and laboratory methods to diagnose the level of iron deficiency is typically time-consuming and costly. In recent years, the use of image processing methods and artificial intelligence models as non-destructive, inexpensive, fast, and accurate approaches have increasingly garnered the attention of researchers. In this study, a method based on artificial neural networks is presented for the automatic classification of peach tree leaves based on the level of iron deficiency.
800 leaf samples were collected from around Urmia County between June and July 2021 for imaging and iron measurements at the Agriculture Faculty of Urmia University. Utilizing the KNN method, images were classified based on active iron (Fe2+) into four deficiency categories: None-, Low-, Moderate-, and Severe- deficiency. Initially, images were transferred from RGB to HSV and Lab* spaces, with statistical features (mean and standard deviation) extracted for analysis. Principal component analysis was applied, and a neural model, including a hidden layer with a sigmoid tangent activation function and an output layer with a linear activation function, was developed. Utilizing 70% of the data for training and 30% for testing and validation, MATLAB 2018 was employed for image processing, feature extraction, and model development, while model efficacy was evaluated using Accuracy, Precision, Recall, and F1-score metrics.
The results, based on 100 iterations, showed that the model utilizing the first 6 Principal Components (PCs) achieved the highest accuracy (89.73% ± 0.54%), precision (89.59% ± 0.57%), recall (89.52% ± 0.51%), and F-measure (89.55% ± 0.54%) among all scenarios. Models with 5 and 6 PCs exhibited enhanced stability considering the accuracy and standard deviation. Consequently, model number 6, with 6 main components, outperformed others in terms of efficiency. Pertaining to the confusion matrix results for the optimal model with test data the "moderate" class was the most challenging to classify, often being misclassified into adjacent deficiency categories i.e., low- and severe- deficiency. Furthermore, the Receiver Operating Characteristic (ROC) results of the optimal model highlighted its proficiency in identifying all categories, obtaining the best performance for classes None, Severe, low, and moderate, in order.
A good correlation was observed between leaf color changes in peach leaf color and the level of active iron, demonstrating that iron deficiency could be detected using digital images and processing the extracted features. Therefore, the findings suggest that the presented model possesses high repeatability and can effectively categorize iron deficiency into four levels: none, low, moderate, and severe. This study, by modeling the color changes of peach leaves and active iron in the peach leaf, lays the foundation for presenting an efficient, cost-effective, and accurate method to replace conventional laboratory methods for detecting iron deficiency levels.
)