Classification of Weed and Corn Seedling Using Hybrid MobileNetV2 and Support Vector Machine

Classification of weed and corn seedling is important to support weed control and optimal growth of corn plants. Currently, the weed removal process is still done manually by farmers. This manual method makes farmers tired if the cultivation area is large. Besides that, the use of herbicides that farmers spray is not precise on weed, so it negatively impacts the environment and cultivated plants. Therefore, we need a precise weed detection system to correctly classify weed and corn seedlings. The challenge in this research is that the plants in the dataset used are still small, even almost the same size. In this research, the classification method proposed is transfer learning and machine learning. First, MobileNetV2 is used in the feature extraction stage. Then at the classification stage, we evaluate using the Support Vector Machine, Random Forest, and K-Nearest Neighbor methods. This research also tested the accuracy of results with or without segmentation. The experimental results show that hybrid MobileNetV2-SVM without segmentation can achieve 0.985 accuracy. This accuracy is quite precise for the classification of weed and corn seedlings. Future research can apply this method to herbicide tools for spraying weed.