Many global and environmental applications require land-use and land-cover information. There are many land-cover classifications using remote-sensing images, as they are excellent classifiers. However, this requires much training data and at least the same amount of test data. In addition, there are few classification groups because it is difficult to obtain a large amount of training data for each class. Therefore, it is also difficult to use biomass resources of unknown forests. In traditional models, so much manpower, money, and time are required to conduct field research. This study classified remote-sensing imagery for different time series by semi-supervised learning based on the maximum-likelihood method and found the location that matched other classification results. In addition, this method could classify fine and large land cover with one training sample. This paper demonstrated the feasibility of biomass resource utilization and estimated resource amounts and position, discussed the transport scenario accurately for using Google Map, and evaluated operation cost for locally produced and consumed woody biomass resources. A recursive maximum-likelihood method is proposed to consult boosting approach. It provided good performance and presented a scenario for using woody biomass resources. In future work, it will be necessary to broaden the objective area and find a relationship between vegetation condition and capital potential. On that basis, we need to research a scenario of locally produced and consumed woody biomass resources for a broad area. 1