Radish Plant Growth Monitoring Using Multimodal Fusion

In the realm of precision agriculture, successful plant growth monitoring plays a pivotal role, and it is greatly aided by advanced deep learning techniques utilizing data from various sensors. The key objective of the paper is to develop a model to monitor the growth of the radish plant through the integration of multi-modal data, specifically RGB and Depth (RGB-D) images. Plant growth can be modeled based on many parameters, for example, the height of the plant, the environmental conditions (such as temperature, humidity, and light intensity), nutrient levels in the soil, leaf count, and the presence of pests or diseases, etc. The proposed study estimates the age of the radish plant with an end-to-end deep-learning model. We developed a Fused Image Transformer (FIT) model to estimate the age of the plant in weeks. The FIT uses a self-attention mechanism to estimate the essential features of the images which will help to complete the objective. The proposed method is compared with the state-of-the-art models for regression problems. We have also validated the FIT model with or without depth information. The Mean squared loss with and without depth information is found to be 0.025 weeks and 2.059 weeks respectively which shows a significant improvement by using the depth information.