Maturity Stage Prediction for Cabbage Harvesting Using Deep Learning Algorithms

Accurate and timely detection of crop maturity is crucial for maximizing the benefits of crop harvesting at optimal time. Estimating crop maturity helps farmers to harvest crop at the optimal time, and to get better quality of the produce. This study investigates the application of deep learning techniques to predict the optimal maturity stage of cabbage for harvesting. The Teachable Machine learning model available on Google was used to detect the maturity of cabbage for harvesting. This algorithm relies on RGB (red, green, and blue) images to classify cabbage into two distinct stages: ‘matured’ (class 1) and ‘pre-matured’ (class 2). A dataset consisting of 630 RGB images were collected from an experimental field using an RGB camera. The Teachable Machine randomly divided this image dataset into two segments: a training set (85%) and a testing set (15%). In this study, the model was trained and tested for three batch sizes, i.e. 16, 32 and 64, three epochs, i.e., 25, 50 and 75, and three learning rates, i.e., 0.01, 0.005 and 0.001. The combination of 16 batch sizes, 75 epochs and a 0.001 learning rate has given the best classification accuracy (95%) for both classes. The identified deep learning architecture was able to classify cabbages as mature and immature with 94% accuracy. The sensitivity (96%) was found higher than the accuracy, which is a good indicator of the performance of the model architecture. The type-II error of developed architecture was only 0.04, which is a better achievement. However, real-world implementation of the model may face challenges, such as varying lighting and environmental conditions, which could affect the model’s accuracy. Addressing these factors will be essential to ensure reliable performance in diverse agricultural settings. Furthermore, model’s use can potentially provide economic benefits to farmers by optimizing harvesting time, which could lead to cost savings and improved yield quality.