Sweet Diseases of the sweet orange leaf cause a significant limitation on citrus production that results in a significant loss of both yield and fruit quality. To help solve this problem, a total of 5,813 pictures were obtained in orchards in Khemerdia, Bheramara, and Kushtia, Bangladesh, in natural lighting, in eleven disease classes. The data was well-balanced and pre-processed to provide consistency and modeling appropriateness. SO-EnsNet is a weighted ensemble model that combines three high-performing CNN architectures, namely DenseNet201, Inception-ResNetV2, and VGG19, whose prediction outputs are fused together using a weighted approach to enhance the overall classification performance. SO-EnsNet was experimentally evaluated and demonstrated a high macro-averaged precision, recall, and F1-score of 98.00% and outperformed all the single models. The suggested method shows consistent detection of various types of diseases, which leads to its strength in real-life agricultural applications. These findings suggest that SO-EnsNet can aid timely intervention and decision-making on crop management, which is a resourceful tool, farmers and agronomists to ensure that the quality of the yields and the losses because of disease outbreaks in the cultivation of sweet oranges are reduced.