A revisit of Internet of Things Technologies for Monitoring and Control Strategies in Smart Agriculture

2022
,
12
, 127
15 of 21
Table 4.
Comparative analysis of different methods based on smart agriculture.
Ref.
Technique Dataset
Disease Classes
Accuracy Sensitivity
Precision
Recall
F1 Mea-
sure
G Mea-
sure
[
44
]
DCNN
[VGG +
AlexNet]
plant
village
dataset
5 (apple rust, scab
black rot, cherry
powdery mildew, and
peach bacterial spots)
97.8%
97.6%
97.6%
-
-
97.6%
[
46
]
DCNN
[en-
hanced
VGG16]
2446
apple
leaves
4 (apple leaf diseases
(Scab, frogeye spots,
cedar rust and healthy)
99.01%
-
99.02%
99.02%
99.02%
-
[
47
]
enhanced
CNN
[AlexNet]
Enhanced
Plant
Village
52
99.11%
-
99.49%
99.11%
99.29%
-
[
48
]
IoT
[WSNs
+ ML
algo-
rithms]
Different
data
-
81.6%
-
-
-
-
-
[
49
]
AlexNet
deep
learning
algo-
rithm
54,306
images
14 crop species and 26
diseases
97.38%
-
97.42%
97.37%
97.36%
[
50
]
LeNet
DL tech-
nique
(X-
Fideo)
PlantVillage
3
98.60%
-
98.82%
97.18%
96.89%
-
For the recognition and detection of olive diseases, such as peacock spot, anthracnose,
and canker, an improved convolutional neural network (CNN) dubbed AlexNet was
suggested. Several innovations separate the proposed model from others. It uses effective
intelligent data preprocessing with a stable image in each class, a transfer learning approach,
and an extended and upgraded PlantVillage dataset to work in more complicated situations.
The total accuracy of the suggested technique is 99.11%, which is the best possible score.
Furthermore, it possesses precision, recall, and F1 measures of 99.49%, 99.11%, and 99.29%,
respectively. Despite the fact that model training takes a long time, classification during
testing takes only a few seconds on a CPU [
47
]. Citrus fruits, leaves, and stems are included
in the image dataset. The collection contains images of normal and diseased citrus leaves
and fruits, including greening, scab, blackspot, canker, and melanosis. There are 759 images
of normal and abnormal citrus leaves and fruits in the data collection. The images had
a resolution of 5202
×
3465 (Mpix), and when scaled at 72 dpi, the width and height
were 256
×
256 pixels, correspondingly. The contaminated images were divided into four
various citrus illnesses and left on their own. The entire process consists of four major
steps: (a) enhancing the dataset using Top-hat and then Gaussian functions; (b) weighted
segmentation and segmentation of lesion through a saliency map, which highlights the
infested area; (c) color, texture, and geometric feature extraction from the diseased area;
and (d) PCA, skewness, and entropy-based feature selection and implementation.
Agriculture management, water contamination, and air quality analysis monitoring
systems were all investigated as part of the smart environment monitoring (SEM) system.
Figure
5
demonstrates that substantial investigation of smart environment monitoring
has increased over the period in both cases, specifically research involving the wireless
sensor network and Internet of Things along with research involving machine learning and
Internet of Things [
41
,
51
].

---

Agronomy