The detection of Fall Armyworm (FAW) infestation in sorghum crop is crucial for effective crop management and minimizing economic losses. Traditional methods of pest detection are time-consuming and often lack precision. There is a need for an efficient, accurate, and scalable solution that can provide farmers with timely information about the presence and severity of FAW infestations to enable better crop management decisions. This study introduces a novel deep-learning solution utilizing the ResNet50 architecture to address the challenge of detecting Fall Armyworm (FAW) infestations in sorghum crop. By training on a diverse dataset of pest images, the system learns to accurately distinguish between healthy sorghum leaves and those infected by FAW. When users submit images of sorghum leaves, the system provides a clear classification, identifying the presence of FAW and predicting the severity of the infection as low, moderate, or highly severe. The main contributions of proposed work are the creation of a sorghum leaf dataset, collected from agricultural farms due to the lack of an existing benchmark dataset, to distinguish between healthy and FAW-infected sorghum with 86.3 percent accuracy and find the severity of infection as low, moderate and high. The results and the innovative approach have been well-received by agricultural scientists, highlighting the practical applicability and impact of the proposed solution. This innovative approach enables prompt detection and provides farmers with essential insights into the extent of the damage, facilitating informed decisions for effective crop management. The integration of severity prediction using ResNet50 enhances the detail and accuracy of the system, offering a significant advancement in pest detection and mitigation strategies, with the potential to improve agricultural outcomes and socio-economic stability.