In the Philippines, there is a growing need for the protection of banana plantation from various diseases that directly affects the livelihood of farmers, markets and overall ecosystem. One such fatal disease is Fusarium oxysporum cubense (TR4 Chlamydospores) which allows growth of such fungi in banana crops that permanently damages the soil for further fertility. As of this writing, there is very small visual distinction between TR4 Chlamydospores and non-infectious Chlamydospores. This paper proposes the use of autoencoders to engineer relevant features in order to distinguish Fusarium Oxysporum from similar fungi or other artifacts present in the soil. Furthermore, the paper tries to address the problem with minimal data available for supervised learning as opposed to traditional methods that require thousands of data points for classification. The purpose of the experiments presented here will aid towards the creation of more sophisticated models to visually discriminate Fusarium Oxysporum.