In Egypt, Cucumber is a crucial cash crop, and its farming could significantly benefit the country's agriculture-based economy. Meanwhile plant disease detection manually is costly and time consuming. This study aims to improve early identification of downy and powdery mildew diseases using machine vision by comparing the effectiveness of several detection methods and developing a real life application. This approach will involve five steps including: image acquisition, pre-processing, feature extraction, post-processing, and classification. In which the experiment was conducted on two greenhouses where 931 images were obtained and used with five key features to train and evaluate the proposed methods. The classification performance of three machine learning algorithms, named discriminant analysis (DA), support vector machine (SVM) and K-nearest neighbors (KNN), were compared. The results indicated that the fine gaussian SVM achieved the highest classification accuracy rate of 96%, where fine KNN got 95.8%, and quadratic DA obtained the lowest value 92.8%. Additionally, the suggested method has a practical application that enables automatic mildew disease detection via personal computers, eliminating the need for sample collection and laboratory analysis. This method could also be extended to identify other plant diseases and pests and track disease progression as the study moves forward.