Fruit Detection and Yield Estimation Using Data Driven Methods

Abstract

Agriculture is the backbone of every agrarian economy country. Due to the increase in population, the demand for food is also increasing day by day, and to fulfill that food requirement, conventional agriculture methods are to be replaced by fast, accurate, and efficient methods to help the nation's demand and increase exports. From this perspective, fruit export plays a vital role in agriculture exports and for that fruit counting is a time consuming and hectic process for cultivators to predict production estimation which not only requires labor concentration but also extensive money and time. This process can be catered to by an accurate, fast, and efficient automated system for fruit counting and yield estimation. In this research study, an algorithm is developed, which detects on-tree fruit in farms, counts them, and predicts the yield estimation for farmers to streamline their harvest process. It is implemented in real-time by capturing the images using a webcam via Raspberry Pi and comparing the results with manual counting and the previous work done in this field. Algorithm includes K-means clustering for cluster differentiation based on colors, the color threshold in HSV color space to filter out only the fruit part that is orange in the present study, and then watershed segmentation to separate clustered fruits. This algorithm was tested on a dataset of 100 images and then in real-time on sample plant (oranges) of different lengths, in different light conditions, and compute the average execution time to process one image. A combined accuracy of 91.23% and an R 2 value of 0.95 was achieved, while the combined average execution time was 1.03 seconds which showed the algorithm is not only fast but also accurate and efficient.