A Soybean Classification Method Based on Data Balance and Deep Learning

Soybean is a type of food crop with economic benefits. Whether they are damaged or not directly affects the survival and nutritional value soybean plants. In machine learning, unbalanced data represent major factor affecting learning efficiency, refer to category in which number samples one much larger than that other, biases classification results towards large thus accuracy. Therefore, effectiveness data-balancing method based on convolutional neural network investigated this paper, two balancing methods used expand set using over-sampling loss function assignable class weights. At same time, verify method, four networks introduced for control experiments. The experimental show new can effectively improve accuracy ability, DenseNet reach 98.48%, but will be greatly reduced by data-augmentation method. With binary use sets, excessive convolution layers lead reduction small purposes. It verified layer 1.52%