5. Discussion and Analysis
It is estimated that plant diseases are a significant contributor to global financial
deficits. Numerous abiotic and biotic stresses and continual tension monitoring concerns
the impacts of the loss of fruit-producing plants. Consequently, the $15 billion U.S. apple
Agronomy
2022
,
12
, 127
14 of 21
industry loses millions of dollars every year. Fruits are one of the most significant sources
of nutrients in plants; yet, illnesses, pests, fungous, infectious, and microbial diseases all
affect the quality and quantity of fruits. Using computer vision-based methods, the issue
may now be alleviated. Diseases/infections may be detected early and effectively using
these methods.
The sickness classification of various fruit leaves was achieved using a deep convolu-
tional neural network DCNN approach. The deep features are retrieved by first utilizing
deep learning networks, such as AlexNet and VGG-s, and then tweaked using a transfer
learning approach. Before the selection step, a multi-level fusion strategy is offered, and
the chosen features analyzed to produce the entropy basis features. To categorize the
obtained feature map, we employed a multi-SVM classifier. The diseases investigated in
the experiments include apple rust, scab, black rot, peach bacterial spots, and cherry pow-
dery mildew, and they were all gathered from a plant village dataset. The recommended
method’s better performance in terms of a 97.8% accuracy, 97.6% sensitivity, 97.6% precision,
and G-measure was observed in the classification results (97.6%) [
44
]. Some research has
investigated whether computer vision approaches may be employed for scalable and early
plant sickness detection. There is still a critical lack of non-lab data sets that can be utilized
to allow vision-based plant disease detection. For visual plant disease identification, the
PlantDoc dataset was supplied. The collection has 2598 data points in total, encompassing
13 plant species and up to 17 disease categories, and was developed by annotating internet-
scraped photos for 300 human hours. Three models for plant disease classification were
trained to illustrate the dataset’s effectiveness. The findings demonstrate that employing
our dataset models may enhance the recognition rate by up to 31%. The recommended
dataset, we feel, will contribute to decreasing the entry barrier for computer vision algo-
rithms in plant disease detection. For photos featuring leaves from various classes in a
dataset with contextual noise, and low-resolution leaf images, the model fails to give proper
conclusions. Using image segmentation methods to extract leaves from the photos can
boost the dataset’s utility. Although the dataset has been rigorously verified, particular
photographs in the collection may be wrongly labeled owing to a lack of sufficient topic
knowledge [
45
].
It is necessary to construct an improved VGG16-based DCNN model to detect apple
leaf diseases (scab, frogeye spots, and cedar rust). The global average pooling layer re-
places the fully connected layer to lessen restrictions and a batch normalization layer is
attached to boost the model’s computational performance. Furthermore, to avoid a long
training time, a transfer learning approach is applied. To detect apple leaf diseases, the
suggested model makes use of 2446 apple leaves, 2141 photos in the training phase and 305
images in the testing phase. The experimental data reveal that utilizing the recommended
approach, the total accuracy of apple leaf classification may reach 99.01%. Furthermore,
the findings demonstrate that cedar rust is accurately diagnosed, but one healthy person is
misclassified as scab and the other as frogeye spots.
Furthermore, the model parameters are cut by 89% compared to the standard VGG16.
As a result, the classification performance is raised by 6.3%, and the computational com-
plexity is cut to 0.56% of the innovative model. Consequently, the DCNN model developed
in this study provides a more accurate and speedier way for recognizing apple leaf infec-
tions [
46
]. Table
4
compares the efficiency of several smart agricultural techniques.
Agronomy
Dostları ilə paylaş: |