The oil palm is an important agro-industrial commodity in Colombia and this kind of crops is severely affected by different diseases. Specifically, the But Rot is the most common disease in oil palm crops. Traditionally, the diagnosis and control of diseases in oil palm crops is an invasive process that requires a time-consuming analysis. Therefore, an automatic classification task is required for identifying this kind of disease. On one hand, hyperspectral images (HSI) are used as a remote sensing tool that allows the identification of different features in a land cover, including diseases in crops and materials. However, the HSI classification is a challenging task, due to the spectral signature is typically associated with a unique class, and consequently, it causes low accuracy on classification results. In other words, most of HSI classification methods do not consider the contributions of one or more materials in the measured spectral pixel, leading to a multiclass model at sub-pixel levels. To address this limitation, spectral unmixing (SU) approach has been used to estimate the contributions of the different materials covered by a pixel. On the other hand, convolutional neural networks (CNN) have demonstrated great a performance in classification tasks using HSI. In this work, a classification approach of the But Rot disease in oil palm is proposed by using SU and CNN through HSI. The architecture of the proposed classifier contains 5 layers which are two convolutional layers, two ReLu layer, and one full connected layer. Simulations results show that the proposed classification method exhibits the best classification performance in terms of the overall accuracy, achieving up 88:7861% using only 20% of the training samples, compared to traditional classification approaches.