Al-based plant Pathology

Summary

The combination of continuous population growth and the gradual desertification of areas of the Earth due to climate change will lead to the inability to provide food for all people in the future, so in order to help the food production to increase we have to find solutions. One solution is to help farmers to detect very fast if their crop is diseased in order to take action before it's too late. Given that in mind this thesis will study the literature and develop an AI-based approach for recognizing anomalies in the canopy, resulting from different diseases. The goal is to support actuation decisions e.g. whether to spray a plant for a given disease. Given that there is a limited amount of properly annotated ground truth data, any solution cannot rely solely on deep architectures that need relatively huge training datasets. Instead the thesis will examine the use of a multi-model approach, where a set of components, each using different machine learning methodologies such as transfer learning, uses knowledge gained from other domains. More specifically, the first component will be an object detection algorithm which recognizes if there is a “leaf” inside an image and as second component will be an algorithm that decides for every leaf, found from the previous component, if it is healthy or diseased and if the leaf found to be diseased it categorizes it among three common leaf diseases. For simplicity reasons the first component will be referred to as CP1 and the second as CP2.