Design and Development of Seed Germination Classification System using AI and IoT

Abstract

In the field of agriculture, seeds play a crucial role since they enable the domestication of a wide variety of plants and the creation of technical developments in the field of biotech nology. The quality of seeds, including their physical, genetic and physiological attributes, is crucial for their expression and impact on agricultural processes. High physiological quality seeds have influenced the development of major crops and have been essential in meeting the demands of a growing world population. Germination testing is crucial for evaluating seed quality, but it may be challenging. Precise environmental conditions are necessary for the procedure, and they must be carefully regulated to guarantee precision. The germination test f indings might be influenced by temperature and humidity. Existing growth chambers are used to maintain a uniform temperature and humidity level, facilitating the germination of seeds. The prolonged time frame for germination testing, which may span from several days to weeks, is an additional concern, possibly causing delays in making germination testing. The manual examination in seed testing laboratories, the limited number of quality testing facilities makes the time limitation even more severe, creating a bottleneck that makes it im possible to scale up and fill the demand for rapid testing and agricultural decision making. The proposed system is an automated growth chamber using Artificial Intelligence (AI) and Internet of Things (IoT) that will precisely regulate temperature and humidity to ensure optimal seed germination. The purpose of this method is to enhance the accuracy of pre dicting seed germination, hence facilitating the evaluation of seed quality, thereby enhancing agricultural production. Proposed system also assists in collection of large datasets. These datasets are essential for developing and improving AI models that predict seed germination. Anovel two- stage network is developed that uses multiple Convolutional Neural Networks (CNN)to automate seed recognition and evaluation of germination condition evaluation. The Detectron2 framework is employed in the first stage for instantaneous seed segmentation, and this Region of Interest (RoI) is then fed into the proposed CNN model for germination pre diction in the second stage with an accuracy of 84%. To increase the efficiency a novel seed segmentation and classification model uses U-Net and CNN architectures were used. iv The suggested method analyses seed germination using U- Net’s segmentation of images and CNN’s classification. The suggested fusion model has 0.91 pixel accuracy, 0.84 IoU, and 0.90 precision. Seed Encoding and Decoding Network (SeED-Net) achieves 0.94 ac curacy with proposed encoding and decoding segmentation methods. Although the model has considerable analytical capabilities, model weights around 1.8 MB of size are suited for implementation on the Jetson Nano embedded GPU. This technology provides significant advantages to farmers by allowing them to assess and evaluate the germination viability of seeds prior to the planting season.