Paper | Year | Objectives | Algorithms | Data used | Results |
[10] | 2018 | Identification of leaf diseases in Mango plant species. | CNN with three hidden layers inspired by VGG-16, VGG-19 and Alexnet. | Dataset consisting of 1200 mango leaf images (captured by a digital camera) of diseased and healthy leaves. | 96.67% |
[7] | 2019 | Automatic and an early diagnosis of a disease and its severity in mango leaves. | Multilayer Convolutional Neural Network (MCNN) based on AlexNet architecture. | Dataset of 1070 mango tree leaves images captured on real-time and 1130 images from PlantVillage Dataset. | Accuracy: 97.13%; Missing Report rate: 2.87; False report rate: 0 |
[2] | 2019 | Detection and identification of mango leaf diseases. | ResNet-CNN (ResNet18, ResNet34 and ResNet50) + Transfer Learning. | Original mango dataset comprises 8853 images captured infield; and Mango leaflet dataset of 8852 images after data augmentation. | ResNet18: 91%; ResNet34: 90.88% and ResNet50: 91.50% |
[25] | 2019 | Defects detection on the surface of mango fruit. | LabView. | 180 mango fruits images of two indian categories (Chausa and Dashehari) with various degrees of defects acquired using color (RGB) camera. | accuracy: 88.6% efficiency: 93.3% |
[3] | 2020 | Mangoes quality grading. | The deep learning models selected for our classification task are AlexNet, VGGs, and ResNets + Transfer Learning. | Dataset of 6400 images (AICUP2020) of single mangos each labeled with a quality grade of either A, B, or C based on the evenness of color and severity of defect or diseases. | accuracy: 83.5% |
[6] | 2020 | Detection of early disease on plant leaves with small disease blobs, which can only be detected with higher resolution images. | ANN (Feed-Forward Neural Network) and Hybrid Metaheuristic Feature Selection in comparison with CNN models (AlexNet, VGG16, ResNet-50). | Dataset of 450 images of mango leaves, which belong to four different types (three diseases and one healthy). The images are captured using a camera in the resolution of 3096 × 3096 pixels with no background under different lighting conditions in a chamber. | FFNN proposed: 89.41% |