TY  - JOUR
AU  - Jepkoech, Jennifer 
AU  - Kenduiywo, Benson Kipkemboi 
AU  - Mugo, David Muchangi 
AU  - Tool, Edna Chebet 
PY  - 2022
TI  - A Backward Regressed Capsule Neural Network for Plant Leaf Disease Detection
JF  - Journal of Computer Science
VL  - 18
IS  - 9
DO  - 10.3844/jcssp.2022.821.831
UR  - https://thescipub.com/abstract/jcssp.2022.821.831
AB  - This study investigated the introduction ofbackward regression coupled with DenseNet features in a Capsule Neural Network(CapsNet) for plant leaf disease classification. Plant diseases are consideredone of the main factors influencing food production and therefore fast cropdisease detection and recognition are important in enhancing food securityinterventions. CapsNets have successfully been adopted for plant leaf diseaseclassification however, backpropagation of signals to preceding layers is stilla challenge due to low gradient flow. In addition, parameter and computationalcomplexities exist due to complex features. Therefore, this study implemented aloop connectivity pattern to improve gradient flow in the convolution layer andbackward regression for feature selection. We observed a 99.7% F1 score withbackward regression and 87% F1 score without backward regression accuracy ontesting our framework based on the standard Plant Village (PV) dataset comprisingten tomato classes with 9080 images. Additionally, CapsNet with backwardregression showed relatively higher and stable accuracy when sensitivityanalysis was performed by varying testing and training dataset percentages. Incomparison Support Vector Machines (SVM), Artificial Neural Networks (ANN),AlexNet, ResNet, VGGNet, Inception V3, and VGG 16 deep learning approachesscored 84.5, 88.6, 99.3, 97.87, 99.14, and 98.2%, respectively. These findingsindicate that the introduction of backward regression of features in theCapsNet model may be a decent and, in most cases superior and less expensivealternative for phrase categorization models based on CNNs and RNNs. Therefore,the accuracy of plant disease detection may be enhanced even further with the aidof the fusion of several classifiers and the integration of a backwardregressed capsule neural network.