A DEEP LEARNING MODEL FOR DECISION SUPPORT OF STAKEHOLDERS IN CROP DISEASES AND PEST MANAGEMENT

Abstract

Crop diseases and pest infestations pose a serious threat to food security and the livelihoods of farmers across the world, posing substantial difficulties to the agricultural industry. Stakeholders, such as farmers, agronomists, and agricultural extension officers, need decision support systems that are both accurate and fast to handle these difficulties. A deep learning model tailored to decision support in crop disease and pest management was the intended outcome of this study. The goals were to create a hybrid model for correct crop diseases and pest classification using a deep neural network, create an easy-to-use software platform, validate the model's generalizability in different agricultural contexts, and identify and analyze important factors affecting decision support. The first objective of the research included a thorough examination of the main features of deep learning models that influence decision support for crop diseases and pest management. The collection of high￾resolution, labelled datasets of crop images from the Kaggle website was done with an emphasis on data quality. These datasets represented diverse forms of crop diseases and pest infestations. In order to increase the classification accuracy, a hybrid model was built that combined a Convolution Neural Network (CNN) with Support Vector Machines (SVM). This model leveraged the capabilities of both techniques. Stakeholders were able to upload images of affected crops and get prompt diagnoses and recommendations when the model was integrated into an intuitive software platform. Extensive testing was conducted across several datasets, including tomatoes, maize, and pests, to assess the model's robustness and generalizability. On the maize data set, the hybrid CNN-SVM model achieved 97.13% accuracy, on the tomato dataset, it achieved 97.04% accuracy, while on the pest dataset, it achieved 97.02% accuracy. Analyses comparing DenseNet, AlexNet, EfficientNet, and VGG showed that their accuracy rates were lower; DenseNet achieved 96.99% for maize, 92.60% for tomatoes, and 96.94% for pest classification. Stakeholders were able to improve their capacity for proactive pest and disease management in crops through the use ofthe user-friendly software platform's proactive decision support. Data quality, efficient feature extraction, model design, and environmental conditions consideration were some of the important factors emphasized by the study as impacting decision support in crop diseases and pest management. In order to get reliable predictions, the results highlighted the need to use high-quality training data and complex model architectures. The study proved that deep learning models greatly enhancedagricultural decision-making by addressing problems related to crop diseases and pest management. When it came to decision support for stakeholders dealing with pest control and crop diseases, the developed hybrid model and its incorporation into an intuitive software platform provided substantial improvements. The hybrid model,according to the research, improved agricultural sustainability and production by providing correct insights in a timely manner. To achieve broad acceptance and influence in the agricultural sector, future work should attempt to enhance the software platform, validate the model's performance across several agricultural settings, and increase the datasets. The findings and methods presented here transformed crop diseases and pest management by providing useful information for decision support systems in the agricultural sector.