The purpose of a Controlled Ecological Life Support System (CELSS) is to achieve life support in the extreme environment through the regeneration and circulation of materials. Along with scale expansion of a space habitat, the size of the CELSS will be also expanded. Therefore, the CELSS must be able to deal with system modifications in a flexible way. But so far, there is no procedure that ensures both the expandability and the stability of the overall system control. In this paper, to ensure both abilities, I propose a new hierarchical autonomous control procedure based on automatic scheduling and multi-agent learning control methods. To ensure the stability of control, an overall CELSS circulation control, called upper-layer control, was planned by the Lagrangian decomposition and coordination method. The elements in the subsystems were controlled by the multi-agent learning method that could easily to deal with system modification, to achieve the control plan constructed by the automatic scheduling method. I simulated material circulation with system modification, such as the addition of tanks, processors, and habitation and plantation modules to check the procedure’s expandability and control stability. As a result, the upper-layer control system responded well to the expansion of the CELSS by re-scheduling. An appropriate response to the updated system was also observed in each subsystem. Guaranteeing effective overall control of the CELSS, with a flexible response to system modifications was demonstrated to be possible, which had been difficult under a decentralized autonomous control scheme alone.
It has become an important solution for the modern agriculture to monitor rice plants in paddy field by remote sensing. In recent years, many researchers have employed some low-cost and high-performance UAVs with cameras for this purpose. Especially, the structure from motion (SFM) has been considered as a method of reconstructing a three-dimensional (3D) model by repeatedly calculating a feature projection point for a plurality of images overlapping. In this paper, to use the SFM method, we took many videos of the rice filed by a color video camera mounted on a small UAV and picked a series of still images from the videos at different video sampling rate. And, we found that 3D models of rice plants in paddy field were well reconstructed from the four processing steps of point cloud building, dense point cloud building, mesh modelling, and textured mapping. The result showed that high sampling rate led to high accuracy and 3D dense point cloud model was better in the accuracy than others. In the sampling rate of 6 still images/s, the error of 3D model was RMSE= 12.8 cm (R2 = 1.00) in X-Y axis and RMSE = 7.3 cm (R2 = 0.97) in Z axis.
In the present study, a method was proposed to separate two plant tissues (photosynthetic and nonphotosynthetictissues) on 3-D point cloud data of a two evergreen trees obtained by a dual wavelength portable scanning lidar, that allowsto get refection intensity of red and near infrared of a target. First, 3D point cloud data of the trees were collected fromground positions that surrounded the trees. Next, the data were voxelized and training data that correspond to a part ofphotosynthetic and nonphotosynthetic tissues were picked out from the lidar data. Based on the training data, distributionsof the refection intensity of red and the ratio of the refection intensity of red and near infrared were investigated and theywere used for separation of two tissues in the lidar data based on the maximum likelihood method. As a result, over allaccuracy and kappa coefficient values of the two trees for the separation ranged from 81 to 93% and from 0.30 to 0.63.