As an important natural resource, forest plays a vital role in regulating regional climatic conditions and maintaining the balance of the Earth’s ecosystem. At the same time, the major changes in China’s natural resource management system have put forward new requirements for forest surveys. Efficient and accurate grasp of the spatial distribution of forest resources and the type of forest in China have become an important part of all-weather remote sensing survey and monitoring of natural resources. In recent years, with the strong promotion of the national high-scoring project, the extraction of forest based on high-resolution remote sensing image (HSRI) has become one of the main technical means of forest resource survey in China. However, HSRI contains abundant detailed information and spatial characteristic information, and spectral heterogeneity within the object is increased, which significantly causes difficulties in HSRI processing and information extraction. Therefore, aiming at the problem of extracting forest types from China’s domestic GF-1 satellite image, this paper proposes a classification method based on feature learning using an object-oriented convolution neural network (CNN). First, the study area was under-segmented using the Multi-Resolution Segmentation (MRS) algorithm and referred Estimation of Scale Parameter algorithm to determining the segmented scale. Then, a CNN model was obtained for identifying forest type by parameter adjustment. Finally, the multi-layer feature learning of the CNN and the regional Majority Voting algorithm were combined to classify the forest types within the study region. The overall extraction accuracy of the proposed method in this paper was 0.88. The object-oriented method can avoid the “salt and pepper” phenomenon that exists in the pixel-based classification. CNN can extract the deep features contained in HSRI, which is effectively conductive to category identification. In addition, the method proposed in this paper has a greater applicability to forest extraction from HSRI.