Anthracnose identification in flowering and not in flowering coffee plantations with self-supervised and supervised learning

Antracnosis is a disease that affect the coffee production on Costa Rica farmlands, causing economics losses to farmers. This investigation had used self-supervised and supervised machine learning algorithms to detect anthracnose on drone's images of the farmlands. The machine learning model was pre-Trained using comparative loss function VICReg and the auto-encoder technique. After the embedding generated from the net were reduced by Umap in semi-supervised mode using a small quantity of labels. The resulting reduced embedding was used to classify more images until the labelled dataset was big enough to re-Train the net using supervised methods for precision increase. The final net has a precision of 80% and 20% of the dataset was labelled.