Farmland Scene Classification Based on Convolutional Neural Network

• Published in: 2016 International Conference on Cyberworlds (CW) pp. 159 - 162
• Main Authors: Zhu Deli, Chen Bingqi, Yang Yunong
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.09.2016
• Subjects: CNN; Convolution; deep learning; deep neural network; Feature extraction; Kernel; Machine learning; Neural networks; Neurons; Training
• DOI: 10.1109/CW.2016.33

This paper proposed a farmland scene classification method based on CNN (Convolutional neural network). The farmland image datasets are divided into 4 types, namely, Crops_field, House_field, Not_farming_field and Woods_field. There are 100 pictures in each type, 80 images in each type are used as training sets, and the remaining 20 images are processed as test sets. Design a CNN with 2 convolution layers and 2 sub sample layers.In the training process, input images are restricted to 64*64, and the convolutional kernel is 5*5. Use the opensource toolkit of deep learning namely Tensorflow as the realization platform. After 700 times trainings, we validated the effects on the dataset, The corresponding correct rates of the four scenes are 79%, 82%, 76% and 75%. The result show that this method can achieve satisfactory effect.