Weed detection and classification using deep learning

CNN can solve miscellaneous complications in several applications suchlike manufactures, robotization, sustainable environment, and medical processes. Except for these areas, precision agriculture according to plant crop management is another essential domain. It must detect and restrain weeds in their early growth stages to accomplish plant crop management. Non-chemical weeds control is a crucial aspect of sustainable organic agriculture. This thesis explores the state-of-the-art deep learning algorithm and proposed a novel and simple convolutional network for weeds detection referred to as CovWNET to detect and classify crop and weeds species. Along with this, transfer learning is implemented with conventional benchmark models in order to compare their performance with the proposed model. CovWNET has the second smallest size among the implementations compared in this study. It has roughly 1.5 times more parameters; however, it achieves 2.8% more accuracy as compared to the smallest network, MobileNetV2. Correspondingly, CovWNET has approximately seven times lesser number of parameters, and 1.7% less accuracy compared to the most accurate model of DenseNET. Performance metrics of precision, recall, F1-score, and support, are employed to evaluate the overall performance of each class of the dataset. Keywords: Deep leaning algorithm, Convolutional neural network, CovWNET, Transfer learning architecture, Weed's classification and detection.