Design and Development of a Deep Learning-Based Optimized Framework for Automatic Crop and Weed Classification Using Agricultural Field Image

Weeds significantly reduce the amount of agricultural produce due to their competition with crops for nutrients, water, light and space. Because manual weeding and blanket spraying of herbicides are both labour intensive and time consuming, as well as costly & damaging to the environment, there have been recent advances made in Computer Vision & Deep Learning that provide a method for automating all aspects of farming. The proposed methodology described in this paper details an Optimized Framework based on Deep Learning for the Automated Classification of Crops/Weeds using pictures of Agriculture fields. The proposed methodology categorizes images and subsequently utilises Convolutional Neural Network (CNN's) to extract very deep discriminative features from these pictures and after that uses an Optimised Machine Learning Classifier to classify these features. The performance of the proposed methodology has been evaluated using Accuracy, Precision, Recall and F1 Score, with experimental results showing that the framework reached 97.90% Accuracy, 98.35% Precision, 97.54% Recall and 97.94% F1 Score, indicating that it works effectively and is Robust and can be applied to Smart Agriculture Applications in a Real-Time basis - Robotic Weed Removal, Precision Spraying and Automated Crop Monitoring.