|
Precision agriculture using artificial intelligence to detect and classify nitrogen deficiency in rice leaves |
|---|---|
| รหัสดีโอไอ | |
| Creator | Suaree Nakornpan |
| Title | Precision agriculture using artificial intelligence to detect and classify nitrogen deficiency in rice leaves |
| Contributor | Udon Jitjuk, Kanoklada Taothaichana, Soonthorn Choksawatthanakit |
| Publisher | Faculty of Agriculture |
| Publication Year | 2569 |
| Journal Title | Khon Kaen Agriculture Journal |
| Journal Vol. | 54 |
| Journal No. | 1 |
| Page no. | 93-106 |
| Keyword | Precision agriculture, artificial intelligence, detection and classification, nitrogen deficiency in rice leaves, CiRACORE |
| URL Website | https://li01.tci-thaijo.org/index.php/agkasetkaj |
| Website title | Khon Kaen Agriculture Journal |
| ISSN | 3027-6497 (Online) |
| Abstract | This research aims to apply artificial intelligence (AI) technology to develop a prototype precision agriculture system for detecting and classifying nitrogen deficiency in rice leaves using the Leaf Color Chart (LCC) from the International Rice Research Institute (IRRI). The system categorizes nitrogen deficiency into four levels: Level 1, severe nitrogen deficiency; Level 2, moderate nitrogen deficiency; Level 3, slight nitrogen deficiency; and Level 4, no nitrogen deficiency. The CiRA CORE program, an artificial intelligence tool, was used to detect and classify nitrogen deficiency in rice leaves trained to recognize the characteristics of rice leaves with all four nitrogen deficiency levels. This program serves as a guideline for determining the appropriate fertilizer application amount for each rice growth stage. In this case, tillering was measured at 40–45 days old. The total number of images used is 600 images, divided into 2 sets, consisting of a training set for training the model, 500 images, divided into 100 images for Detection training, 400 images for Classification training, and 100 images for testing the model's performance after training is complete. The test results showed that the developed AI model is effective in classifying nitrogen deficiency as follows: Level 1 has a precision of 95 percent, a resolution of 92 percent, and an F1-Score of 95 percent. Level 2 has a precision of 92 percent, a resolution of 90 percent, and an F1-Score of 91 percent. Level 3 has a precision of 90 percent, a resolution of 91 percent, and an F1-Score of 91 percent. Level 4 has a precision of 96 percent, a resolution of 97 percent, and an F1-Score of 97 percent. The results of using the agricultural model were evaluated in 4 aspects. It was found that the speed of processing had the highest average value of 4.80, with a standard deviation of 0.30. Farmers were able to reduce the cost of nitrogen fertilizer by 30 percent (374 baht/rai). In conclusion, the developed AI model can help reduce fertilizer application costs in rice fields by efficiently analyzing nitrogen deficiency in rice leaves, aligning with the plant's growth stage and fertilizer application amount. |
