Residual Global Context Assisted Deconvolutional Single Shot Detection Network for Cotton Boll Growth and Harvesting Stage Prediction

Abstract

Cotton plays a vital role in all over the world, and time of reaping is essential to increase the quality of cotton. Anticipating and detecting the stage for harvesting is a problem in cotton plants. Existing methods in predicting the harvesting stage is time-consuming and labour-intensive. This paper introduces a deep learning method to accurately determine the growth stages of cotton. First of all, the images from the dataset are enriched through augmentation to steadiness the dataset and lessen over fitting using Generative Adversarial Network (GAN). After the augmentation, the images undergo pre-processing to reduce noise by "Guided Gaussian filter approach (GGF)". Next, the prediction and reaping stage is identified through advanced deep learning model named "Residual Global Context assisted Deconvolutional Single Shot Detecting Network (GCNN-DSSD)". The Deconvolutional Single Shot Detecting Network architecture employed for the predicting the cotton plant accurately from the cotton field. Residual Global Context Convolutional Network correctly detects the stages of cotton boll, whether it is suitable for harvesting. Experimental outcomes of our suggested method are analysed using python tool and acquired values of Accuracy of 98.76%, precision of 98.23%, Recall of 97.52%, F1-score of 97.46%.