Various types of cylindrical biomass particles (pine, beech, bamboo, demolition wood) have been pyrolyzed in a batch-wise operated fluid bed laboratory setup. Conversion times, product yields, and product compositions were measured as a function of the particle size (0.7-17 mm), the vapor's residence time (0.25-6 s), the position of the biomass particles in the bed (dense bed or splash zone), and the fluid bed temperature (250-800 °C). For pyrolysis temperatures between 450 and 550 °C, the bio-oil yield appeared to be maximal (in this work: about 65 wt %), while the water content of the bio-oil is minimal. The position of the biomass particles in the fluid bed, either in the dense bed or in the splash zone, does not affect the conversion time and product yields to a large extent during pyrolysis at 500 °C. In the small fluid bed used for this work, with a char hold-up of up to 5 vol % (or 0.7 wt %), the residence time of the pyrolysis vapors is not that critical. At typical fast pyrolysis temperatures of around 500 °C, it appeared sufficient to keep this residence time below 5 s to prevent significant secondary cracking of the produced vapors to noncondensable gas. Up to a diameter of 17 mm, the particle size has only a minor effect on the total liquid yield. However, for particles larger than 3 mm, the water content of the produced bio-oil increases significantly. The experimental results are further compared with predictions from a one-dimensional (1D) and a two-dimensional (2D) single-particle pyrolysis model. Such models appeared to have a limited predictive power due to large uncertainties in the kinetics and selectivity of the biomass decomposition. Moreover, the product quality cannot be predicted at all.

We present a new dynamic programming algorithm that solves minimum Steiner tree problems with k terminals in time O * (c k) for any c > 2. This improves the running time of the previously fastest exponential time algorithms (Dreyfus-Wagner [2]) of order O * (3 k) and the so-called "full set dynamic programming" algorithm, cf. [3], solving rectilinear instances in time O * (2.38 k).

Human detection has been playing an increasingly important role in many fields in recent years. Human detection is a still challenging task because, for the group of people, each individual has his unique appearance and, body shape. Compared with the traditional method, the deep learning neural network has the advantages of shorter computing time, higher accuracy and easier operation. Therefore, deep learning method has been widely used in object detection. The current state of art in human detection is RetinaNet. Among all the deep learning approaches, RetinaNet gives the highest accuracy of human detection (Lin, Goyal, Girshick, He, & Piotr Dollar, 2018).The temporal component of video provides additional and significant clues as compared to the static image. In this paper, the temporal relationship of the images is utilized to improve the accuracy of human detection. Compared to using only an image, the accuracy of human detection is 21.4% higher when a sequence of images is applied.