Performance analysis of the application of convolutional neural networks architectures in the agricultural diagnosis

Abstract

Agriculture is an important sector for developing countries and farmers. Recently, numerous techniques for increasing agricultural productivity have been utilized. However, different issues are still encountered by farmers including various plant diseases. Plant diseases diagnoses are challenging research, and they should be analyzed and treated by detecting the diseased plant leaves. For that reason, in this paper, we develop our proposed architecture using convolutional neural networks (OP-CNN) as a computer-aided to detect and diagnose plant diseases. The proposed architecture can assist farmers in increasing both the quantity and quality of their agricultural productivity. Besides this, the OP-CNN helps to reduce disease prevalence through early detection. The performance of our proposed model is compared with other convolutional neural networks (CNN) architectures in order to validate its capability. The strawberry dataset was employed to train and test the models since the strawberry is one of the main crops in the Larache Province (Morocco). The experimental tests demonstrate that our proposed OP-CNN reaches the highest values versus DenseNet121, VGG19, and ResNet50 with 100%, 99%, 97%, and 63% respectively for classification accuracy, 100%, 100%, 98% and, 79% respectively for precision, 100%, 99%, 97%, and 63% respectively for recall, and 100%, 99%, 97%, and 58% respectively for "F" _1Score.

Fine-tuning a pre-trained ResNet50 model to detect distributed denial of service attack

“…Through harnessing this pre-training, the model can be fine-tuned or employed as a feature extractor for a range of computer vision assignments, including image classification, object detection, and image segmentation [29], [30]. There are several CNN-based methods that are widely used for various applications [31]- [33], but in this study, we focus on ResNet50. We will include other deep learning methods on future work agendas.…”

Citrus leaves disease diagnosis

“…Different feature maps can also be applied multiple convolutional layers. This strategy is to assure complete extraction of various features [11], [17]- [19]. Convolutional processing offers three main benefits.…”

“…Feature maps size is then reduced by the pooling layer. This procedure improves the input's resistance to noise and distortion [11], [17]. The pooling layer is increasingly employed to reduce measurements of the network parameters and function mappings.…”

“…The majority of implementations employ max-pooling because it can increase generalization, speed up convergence, and select superior invariant features [3], [20]. CNN depend on 3 rd layer (the activation function) [11], [17]. Convergence of the neural network heavily depends on selecting the suitable activation function.…”