Computer Vision enabled Plant's Health Estimation in Precision Farming

Abstract

Agriculture plays a vital role in blooming the economy of agriculturally rich countries. In Pakistan, agriculture supports the population for 26% of GDP. To enhance the quality and quantity of crops, the use of pesticides is of great importance. The excessive and non-precise use of pesticides can decline the productivity and can also cause serious health hazards to farmers. In this paper, we propose a concept of using computer vision to monitor pesticide usage according to the health of crops. The proposed algorithms will train the machine for the diseased and non diseased crop by using a data set of both features. Based on this, real-time images of the crop are compared to the data set on which it is trained on. Classification of the crops are done via Tensorflow into diseased or non diseased. If the crop turns out to be diseased, the robot automatically sprays pesticide on the selected plant. The highest accuracy, precision, recall and F1 score achieved by our proposed model are 81%, 0.8531, 0.8513 and 0.9522, respectively.