Determining the amount of nitrogen in greenhouse cucumber leaves in order to manage fertilizer consumption by image processing
Subject Areas :
Keywords: image processing, Neural network, Chlorophyll, greenhouse cucumber, leaf nitrogen content,
Abstract :
Majid Hadda M.Sc, Department of Biosystems Engineering, Takestan Branch, Islamic Azad University, Takestan, Iran Abstract The external diagnosis of nutritional disorders may be confused with symptoms caused by non-nutritional factors such as diseases, pests and chemical compounds, therefore leaf analysis can be used to confirm the external diagnosis. With proper nitrogen management, it is possible to provide the right amount of nitrogen to the product and also reduce the pollution caused by it in soil and water. The purpose of this research is to determine the possibility of using digital images to determine the amount of nitrogen in greenhouse cucumber leaves and compare it with the laboratory analysis method. The results obtained from the evaluation of the laboratory method of measuring the chlorophyll of greenhouse cucumber leaves with the Keldal apparatus and the image processing method for 50 healthy and damaged leaves were compared with each other. Then the accuracy of the algorithm designed to determine the percentage of chlorophyll and the damaged part of the plant leaf was evaluated and finally the best neural network was selected as the most accurate model. The results showed that the average amount of chlorophyll in cucumber leaves was approximately 72.65% for the laboratory method and approximately 74.14% for the image processing method, which indicates the low accuracy difference in these two methods and the use of non-destructive image processing method. compared to the laboratory method. The proposed model in order to detect the amount of chlorophyll and the damaged part of the leaf by using the area detection with unsupervised k-means clustering method, has a network with two neurons in the first hidden layer and five neurons in the second hidden layer and the training function and It was a transfer function, and as a result, a network with 5-5-2-15 topology was formed, and 100% accuracy was obtained for this model.
_||_