The Effect of Data Augmentation and Padding of the Image Dataset on Detection of Black Sigatoka Disease on Banana Leaves Using ShuffleNet V2 CNN Architecture

In this study, image processing is paired with Shufflenet V2 CNN Algorithm to detect Black Sigatoka infected banana leaves with the aid of Data Augmentation and Padding for testing and training. Black Sigatoka disease is one of the most important diseases for banana crops since it is responsible for most yield depletion and premature ripening. Currently, some studies tackle the detection of banana crop diseases using image classification with different algorithms for differentiating infected from healthy ones. As for the result, the study has tested Shufflenet V2 Algorithm in detecting Black Sigatoka using distinct training datasets. When applying the algorithm, different setups and alterations can be made for its training dataset to fully enhance the overall performance of the algorithm through Data Augmentation and Padding. Each Data Augmentation and Padding setup is tested for the Shufflenet V2 Algorithm in this study. The results show that with this small dataset, the ShuffleNet V2 CNN algorithm trained to detect Black Sigatoka on banana leaves, augmenting the image dataset and padding every image yielding an accuracy of 95%, specificity of 93.33% and sensitivity of 96.67%.