SUPERVISED MACHINE/DEEP LEARNING TECHNIQUES – A CASE STUDY OF POWDERY MILDEW DETECTION ON THE STRAWBERRY LEAF
Abstract
This research proposed the algorithm, that can detect powdery mildew and give the highest classification accuracy (CA). Three image processing and two machine learning algorithms (artificial neural network; ANN and support vector machine) were used to find the optimal combination for different image resolutions. Also, data augmentation by using a rotation technique was carried out to simulate the real-world field-situation. The results after data augmentation tended to be underfitting due to the added directional parameter. The results showed the highest CA of 94.34% for the combination of speeded-up robust features and ANN at 908×908 image size. To get better performance, six convolutional neural network algorithms were compared after data augmentation. ResNet-50 as the highest CA, AlexNet as the shortest computation time, and SqueezeNet-MOD2 as the smallest memory would be recommended in the conclusion.