<![CDATA[Cocoa is an important commodity in Indonesia that is susceptible to pathogen-induced diseases. These diseases reduce fruit quality and are difficult to recognise at an early stage. This research uses a Convolutional Neural Network (CNN) method with a transfer learning approach and NASNet-Mobile architecture to facilitate the classification of cocoa fruit diseases. The data consisted of 2000 images of diseased and non-diseased cocoa pods divided into four classes, namely Cocoa Pod Rot (Black Pod), Fruit Sucking Ladybugs (Helopeltis sp), Fruit Borer (Pod Borer) and Normal. Training was conducted for 25 epochs using Google Colab. The best model produced 99.11% training accuracy, 96.14% validation, and 94.88% testing. The model was implemented into an Android device and field tested with 93.33% accuracy, 98.5% recall, 57.1% precision, and 71.6% F1-score. This system is effective in helping early detection of cocoa pod disease in a practical, efficient manner without reducing the accuracy value.]]>
Copyrights © 2025
