Abstract:
Due to divease and pest attacks, the agriculture vector is at stake globally. Food is the basic necessity of every human being. Alongside multiple sir al and bacterial di...Show MoreMetadata
Abstract:
Due to divease and pest attacks, the agriculture vector is at stake globally. Food is the basic necessity of every human being. Alongside multiple sir al and bacterial diseases, natural calamines like critical weather conditions, which invoke floods and land sliding, also play a role in crop production quantity and quality degradation. Hence, there is a need to develop precise and automated methods capable of diagnosing and classifying diseases in crops. This article proposes a Deep Learning (DL) based architecture to identify and classify the disease on time and build a lightweight Deep Convolutional Neural Network (DCNN). Wheat and cotton datasets are used for detailed experimentations. InitiaDy. the image is enhanced using three state-of-the-art contrast enhancement techniques: histogram equalization, adaptive histogram equalization, and contrast-limited adaptive histogram equalization. These techniques are used to improve the visual quality of the images and enhance the model's ability to identify disease patterns. A novel Residual Block-biLSTM architecture is used for training. The strength of this network lies in a residual block in concatenation with the Bi-LSTM byer. The residual block helps the model to train intensely by enhancing the gradient flow in the network more smoothly. In contrast, the Bi-LSTM layer (Bidirectional Long Short-Term Memory) is a Recurrent Neural Network layer type, which combines the working capabilities of typical LSTM networks with the additional power of bidirectional processing to extract sequential patterns in the input data. Significant features thus extracted are used for classification. The % accuracy rate of 9S22% is achieved on wheat, and 95.46% is achieved on cotton datasets. Results show improved accuracy with enhanced training accuracy, a lower error rate, and reduced prediction time, making the proposed model lightweight and precise for large and hand-held devices.
Date of Conference: 15-16 October 2024
Date Added to IEEE Xplore: 09 December 2024
ISBN Information: