An integrated OkraNet Framework for detection of disease and maturity stage classification in okra farming

Abstract

Okra (Abelmoschus esculentus) is a vital crop in the Indian agriculture sector, producing one-third of its production. Identifying fresh and ripe okra plants for maximum yield and profit is significantly challenging. Ripeness can be determined by shape, length, color variation, and moisture content. However, to reduce this time-consuming effort, this work emphasizes the classification of fresh and diseased okra leaves as the initial step and assesses the maturity stages, including ripe, unripe, and overripe. The OkraFarm dataset was collected from the real-time farm to determine the maturity stage. Building on state-of-the-art convolutional neural networks, three experiments are performed to lay identification of fresh and ripe okra—Experiment 1: leaf disease classification using the pre-trained VGG19 model achieving a maximum accuracy of 98.89%; Experiment 2: detection of okra fruit using the YOLOv5 model, achieving a maximum accuracy of 84.5%; Experiment 3: handling data imbalance using the MLSMOTE algorithm and classifying the maturity stages of the okra plant into ripe, overripe, and unripe, achieving a maximum accuracy of 96.10% on the test data.