Abstract:
The concept of the hydroponic garden is now prominent in the Sri Lankan non-agricultural community that urged with the Covid-19 Pandemic and economic crisis, where people need fresh and healthy food at a low cost. These people will need help identifying the nutrition deficiencies of plants. They cannot provide the adjustments of nutrition supply to the growing medium, usually the aqueous solvent, which ultimately leads to lower harvest rates. This research suggested an approach to identify the nutrition deficiencies related to diseases for four common plants: Beans, Bell-pepper, Tomatoes, and Gherkin. Calcium deficiency for Tomatoes, Nitrogen deficiency for Bell-pepper, and Magnesium deficiency for Beans and Gherkin based on either an image or video footage. The research utilized the pre-trained Convolutional Neural Network (CNN) model MobileNet for Transfer Learning. TensorFlow Object Detection API with “ssd-MobileNet-coco-v2” is used as the base and specially designed Nutrition Deficiency Algorithm (NDA) in the deficiency identification process. The data set of 1269 video footage of selected plants was annotated by the LabelMg tool. The proposed solution takes an image or video stream, analyses the image, and identifies the nutrition deficiencies. It can draw boxes and give the results, which contain the detected nutrition deficiency and the percentage value of the identification rate. Also, the solution can identify the multiple deficiencies for a single image or video stream. Among the four plants, Magnesium deficiency gives the highest accuracy at 94% for Gherkin, while Calcium deficiency for Tomatoes gives the least accuracy rate at 65%. Moreover, 78% was recorded for the Bell-pepper plant for Nitrogen deficiency, while Magnesium deficiency for the Beans was recorded as 70%. This solution can identify the above nutrition deficiencies and overcome the lower accuracy rate of the human-based deficiency identification process, including subjective errors like fatigue and the need for more expert knowledge.
