Fig 1.
The data flow diagram of the DCDM that illustrates the process of our proposed disease diagnosis.
Fig 2.
Identification & classification of strawberry plant leaf disease by AWS DeepLens in real-time.
Fig 3.
Sample images from dataset: (a). Apple Scab, (b). Black Rot, (c). Cedar Apple Rust, (d). Apple Healthy, (e). Grape Black Rot, (f). Grape Esca, (g). Grape Leaf Blight, (h). Grape Healthy, (i). Peach Bacterial Spot, (j). Peach Healthy, (k). Potato Early Blight, (l). Potato Late Blight, (m). Potato Healthy, (n). Strawberry Leaf Scorch, (o). Strawberry Healthy, (p). Tomato Bacterial Spot, (q). Tomato Early Blight, (r). Tomato Late Blight, (s). Tomato Leaf Mold, (t). Tomato Septoria Leaf Spot, (u). Tomato Spider Mites, (v). Tomato Target Spot, (w). Tomato Leaf Curl Virus, (x). Tomato Mosaic Virus, (y). Tomato Healthy. From PlantVillage: (c), (d), (e), (g), (j), (k), (l), (m), (r), (s), (t), (w) and (z). From Tarnab Farm: (a), (b), (f), (h), (i), (n), (o), (p), (q), (u), (v) and (y).
Fig 4.
Data augmentation technique examples: (a). Original Image, (b). Blur, (c) Random Gaussian Noise, (d). Random Contrast, (e). Random Bright, (f). Scale Proportionality, (g). Random Crop, (h). Deterministic Crop, (i). Vertical Flip, (j). Horizontal Flip, (k). Rotate Without Padding, (l). Y-Sheared.
Table 1.
The dataset for leaf disease classes.
Fig 5.
The representation of DeepLens Classification and Detection Model (DCDM) architecture.
Table 2.
The summary Of DCDM layered architecture.
Table 3.
Hyper-parameters of the experiments.
Fig 6.
Basic workflow of a deployed AWS DeepLens project [55].
Fig 7.
Visualization of feature map from DCDM convolutional layer for a sample leaf.
Fig 8.
Visualisation of filter activation in DCDM convolution layers.
Table 4.
Dataset split for training and testing.
Table 5.
Dataset split for training/testing and accuracy obtained per epoch.
Fig 9.
Trend graph for accuracy and loss in training and validation.
Fig 10.
Confusion matrix for 80-20% dataset split set.
Table 6.
DCDM performance report.
Fig 11.
Sample results from real field and controlled environment images.
Fig 12.
Average accuracy obtained by each CNN model.
Table 7.
Average time consumed by CNN’s per epoch.