An efficient deep learning with a big data-based cotton plant monitoring system

In Agriculture, plant monitoring plays an important role from seedling to harvesting which helps farmers achieve a good yield. This paper focuses on building bigdata based cotton plant monitoring system. To build this system, the plant images are collected from the agricultural field with an Android mobile App. The collected data are labeled using a web-based feature labeling application. After applying the pre-trained Deep-Learning (DL) classification algorithm to the labeled images, the farmers benefited from the subsequent information. This paper discusses different pre-trained Convolutional Neural Network (CNN) architectures such as ResNet18, GoogLeNet, InceptionV3, and MobileNetV3 Large used to monitor the health of the cotton plant. In plant health monitoring, the classification and identification accuracy are improved with better feature extraction. In comparison with other methods employed in this paper, MobileNetV3Large provides high accuracy. The proposed model classifies 11 different cotton plant regions which are boll, bud, crown, flower, land, leaf, stem, unhealthy leaf, weed, young boll, and young leaf. The MobileNetV3Large model offers an accuracy, specificity, and precision value of 93.9%, 96.12%, and 97.48% when evaluated using the images obtained from smartphones. The smart application developed also provides information to the framers regarding harvesting and yield. The proposed model is determined in real-world applications to identify whether a plant sprouted is a cotton plant or a weed. Next, it can also identify the health condition of the cotton plant and can predict the type of disease identified.