Recognition of food images based on transfer learning and ensemble learning.

The recognition of food images is of great significance for nutrition monitoring, food retrieval and food recommendation. However, the accuracy of recognition had not been high enough due to the complex background of food images and the characteristics of small inter-class differences and large intra-class differences. To solve these problems, this paper proposed a food image recognition method based on transfer learning and ensemble learning. Firstly, generic image features were extracted by using the convolutional neural network models (VGG19, ResNet50, MobileNet V2, AlexNet) pre-trained on the ImageNet dataset. Secondly, the 4 pre-trained models were transferred to the food image dataset for model fine-tuning. Finally, different basic learner combination strategies were adopted to establish the ensemble model and classify feature information. In this paper, several kinds of experiments were performed to compare the results of food image recognition between single models and ensemble models on food-11 dataset. The experimental results demonstrated that the accuracy of the ensemble model was the highest, reaching 96.88%, which was superior to any base learner. Therefore, the convolutional neural network model based on transfer learning and ensemble learning has strong learning ability and generalization ability, and it is feasible and practical to apply the method to food image recognition.

Cobalt-catalyzed cyclization of carbon monoxide, imine, and epoxide.

Cobalt-catalyzed cyclization of CO, imine, and epoxide has been developed. A convenient catalyst system composed of Co2(CO)8 and LiCl is identified, and the substrate scope has been explored. The reaction provides an efficient method for the synthesis of substituted 1,3-oxazinan-4-ones.

[Natural regeneration characteristics of Sorbus pohuashanensis in forest region of eastern Northeast China].

Sorbus pohuashanensis is an important non-timber tree species in Northeast China. Aimed to study the natural regeneration characteristics of this tree species and related affecting factors, representative S. pohuashanensis forests in the forest region of eastern Northeast China were investigated by line sampling method. In this forest region, S. pohuashanensis was regenerated by seed propagation, stump sprouting, and root sprouting. In intact or poor habitat natural forests, the proportions of the S. pohuashanensis seedlings established by each of the three regeneration methods occupied roughly a third, with no significant difference (P > 0.05) among them; while in secondary forests, the frequency of stump sprouts (16.5%) was lower than that in natural forests. Even so, the combination of stump sprouting and root sprouting could likely maintain a stable local population. Root sprouting could make the seedlings spread more than 50 cm away from the stump. The transfer rate from diameter class II (1.0-2.9 cm) to diameter class III (3.0-4.9 cm) was 25.6% in secondary forests, 45.3% in poor habitat natural forests, and 15.9% in intact natural forests, suggesting that the lower transfer rate was the key limiting factor for S. pohuashanensis natural regeneration.

[Sorbus pohuashanensis seed dispersal and germination and their relationships with population natural regeneration].

Sorbus pohuashanensis is an important non-timber tree species in Northeast China, but poor in natural regeneration via seed dispersal. In this paper, a field observation was made on the seed dispersal of S. pohuashanensis, its seedling emergence, and soil seed bank, aimed to study the factors affecting the natural regeneration of this tree species. There were 96.1% of naturally dispersed S. pohuashanensis seeds distributed within the range of 2 m around the stock tree, and 97.0% of the seeds in soil seed bank distributed in litter layer and 0-2 cm surface soil. The seed quantity in soil seed bank differed greatly with seasons, being the maximum [(257.7 +/- 69.2) seeds per square meter] in early November in the year of seed dispersal, and the minimum [only (2.9 +/- 2.9) seeds per square meter] in next July. Temperature was not the limiting factor affecting the seedling emergence of S. pohuashanensis [at 0 degrees C-5 degrees C, the seedling emergence percentage could reach (67.5 +/- 6.6)%], but strongly affected the seedling emergence rate. The most proper soil moisture content for the seedling emergence was 50% , under which, the seedling emergence percentage could reach (74.7 +/- 4.2)%. When the soil moisture content was up to 60%, the seedling mortality was the lowest (32.6% +/- 0.6%). All the results suggested that the seed dispersal pattern of S. pohuashanensis and the spatiotemporal dynamics of soil seed bank could affect the seedling emergence of the tree species, and further, affect the population natural regeneration of S. pohuashanensis via seed dispersal.

[Animal predation and dispersal of Sorbus pohuashanensis fruits and seeds].

Sorbus pohuashanensis is an important non-timber tree species in Northeast China. There are two approaches for its fruit- and seed dispersal, i. e. , natural dispersal and animal dispersal. In this paper, a field observation was conducted to study the characteristics of animal predation and dispersal of S. pohuashanensis fruits and seeds, and the effects of the predation and dispersal on the natural regeneration of the tree species. During the fruit-ripening stage of S. pohuashanensis in 2008 and 2009, there were 8 species of birds visiting S. pohuashanensis fruits, among which, three pulp-eating species Turdus naumanni, Cyanopica cyana, and Dendrocopos leucotos could disperse S. pohuashanensis seeds. The visiting frequency of the three species to S. pohuashanensis fruits was 54%, 12%, and 7%, respectively, and the first stop spot after predation was mainly within 5-10 m away from the visited trees (with a proportion of 68.2%), secondly within 5 m (27.3%), and only 4.5% was out of 10 m. The retention time of S. pohuashanensis fruits in the digestive tract of the bird dispersers was up to 20 min, implying a long potential dispersal distance. Fruit- and seed placement tests showed that the fruits artificially placed in different habitats were disappeared within 6-7 days, and the fruit predators on the ground were mainly rodents and ground-feeding birds, with the predation rates being 50% -70%, while the seed predators on the ground were rodents, ground-feeding birds, and ants, with the predation rates only 1% -5%. S. pohuashanensis could provide food for a variety of frugivorous animals, and in return, the animals could disperse S. pohuashanensis seeds, which might have profound effects on the natural regeneration of S. pohuashanensis.