Lugus Lucorum Detection by Improved Convolutional Neural Network

To improve the detection accuracy of Lugus Lucorum in natural environment, a Lugus Lucorum detection method is proposed based on improved convolution neural network (CNN). First, based on YOLO-v3, a new training data set labeling strategy is designed to make the target have a higher effective pixel occupation rate in Ground Truth. Two DenseBlock structures are integrated to effectively alleviate gradient disappearance, reduce the number of parameters, and save computational power. Feature reuse can also play an anti-overfitting role. The proposed method is validated on the dataset of Lugus Lucorum images. The experiment results show that the method can effectively detect the Lugus Lucorum.