Pest Localization and Recognition Using Data Augmentation and CNN Model
In agriculture, pest always causes major damage in the fields and results in significant crop yield losses. Currently, available pest classifications are time-consuming that affectsaccuracy. The existing pest localization and recognition methods based on Convolutional Neural Network (CNN) are not effective for practical pest prevention in fields because of different scales and attitudes of pest. In order to make this localization and recognition more effective, Data Augmentation strategy for CNN-based method is proposed, by rotating the images by various degrees and cropping into different grids data augmentation is applied. The dataset contains the experimental results on various pests affecting the plants. Manual recognition and counting are very time-consuming and inefficient. Automatic pest localization and recognition in the field can reduce the manual work and improve efficiency. All the other methods work to identify the pest in simple backgrounds rather than complex backgrounds of the field environment. Normally while changing the image scales and attitudes extra noise is added to the dataset affect the speed of the training dataset. In order to avoid this problem, a CNN-model is used along with Data Augmentation. A CNN model proposed here is GoogLENet model which works on complex backgrounds and improves the accuracy.