Detection of maturity stages of coconuts in complex background using Faster R-CNN model

Abstract

Coconuts are commonly harvested by judging their maturity based on colour, shape, timeframe, shaking sound, and other growth characteristics of changes as they grow. Currently, solutions involving image-processing techniques have substantial challenges involving the identification of the maturity stages of coconuts. Accordingly, an improved faster region-based convolutional neural network (Faster R–CNN) model is proposed for the detection of two important maturity stages for coconuts in complex backgrounds. The detection of the maturation stages of coconuts for harvesting without human intervention involves challenges because of the complexity of the environment and the similarity between fruits and their backgrounds. Images of coconut and mature coconut bunches were collected from coconut farms. These images were augmented using rotation and colour transformation techniques. These augmented images were used along with original images during model training. The Faster R–CNN algorithm with the ResNet-50 network was used to enhance the detection score of nuts with two major maturity stages. Following training, the detection performance was tested with a dataset that included real-time images as well as Google images. The test results showed that the detection performance achieved using the improved Faster R–CNN model was greater than that for other object detectors such as the single shot detector (SSD) you only look once (YOLO-V3) and Region-based Fully Convolutional Networks (R–FCN). The promising results obtained from this study provided the motivation to develop an application tool for detecting coconut maturity from real-time images on farms.