Machine Learning Methods for Determining Optimal Irrigation Timing for Corn
- 15 October 2024
- journal article
- Published by YUR-VAK in Computational nanotechnology
- Vol. 11 (5)
Abstract
The global forecast for increasing food production on irrigated lands poses the task of optimizing irrigation. Saving water resources is especially important in arid areas, where it is very important to clearly understand what to water, when and in what quantity. The article proposes a method for optimizing the irrigation process of agricultural crops using a control system based on visible and hyperspectral images. We proposed an algorithm and developed a system for obtaining a map of corn irrigation in the low-delay mode. The system can be installed on a circular sprinkler and consists of 8 IP cameras connected to a video recorder connected to a laptop and a hyperspectral camera synchronized with one of the IP cameras. The algorithm for establishing irrigation rates consists of three stages. The stage of establishing the average stage of plant growth (a site of 6–8 plants), the stage of determining the amount of water in plants on this site and the stage of establishing plant irrigation rates directly. In the first case, we used a modified DenseNet121 convolutional neural network with a compression and excitation (SE) block, trained on visible images from an IP camera and allowing to identify the growth stage according to the Biologische Bundesanstalt, Bundessortenamt und CHemische Industrie (BBCH) scale with an accuracy of up to 92%. In the second case, we used hyperspectral images, which, together with the data on the development stage, determine the amount of water in plants. Hyperspectral images were converted into a 2D-model using wavelet transforms and then classified using the 2D-CapsNet capsule neural network. The accuracy of detecting a lack or excess of water in plants was 94%. At the third stage, the data obtained from the two previous stages and a number of characteristics related to the current state of the atmosphere and the field were combined into a separate classifier based on a neural network – a multilayer perceptron, which marked the areas of the field with increased and decreased irrigation rates. The resulting map was then used to irrigate the field. This reduced the amount of water used by 7.4%. At the same time, the efficiency of irrigation water use, linked to the yield of agricultural crops per unit of water used, increased due to an increase in yield by 8.4%.Keywords
This publication has 56 references indexed in Scilit:
- Machine learning methods for the industrial robotic systems securityJournal of Computer Virology and Hacking Techniques, 2023
- Development of Approaches to the Intellectualization of Irrigation Control Systems2021
- YOLOv5 versus YOLOv3 for Apple Detection2021
- Detecting Apples in Orchards Using YOLOv32020
- LCIS DSS—An irrigation supporting system for water use efficiency improvement in precision agriculture: A maize case studyAgricultural Systems, 2019
- A comprehensive review on automation in agriculture using artificial intelligenceArtificial Intelligence in Agriculture, 2019
- IoT and agriculture data analysis for smart farmComputers and Electronics in Agriculture, 2019
- An IoT based smart irrigation management system using Machine learning and open source technologiesComputers and Electronics in Agriculture, 2018
- Adapting weather conditions based IoT enabled smart irrigation technique in precision agriculture mechanismsNeural Computing and Applications, 2018
- Water, Agriculture and Food: Challenges and IssuesWater Resources Management, 2017