A few-shot learning method for tobacco abnormality identification.

Tobacco is a valuable crop, but its disease identification is rarely involved in existing works. In this work, we use few-shot learning (FSL) to identify abnormalities in tobacco. FSL is a solution for the data deficiency that has been an obstacle to using deep learning. However, weak feature representation caused by limited data is still a challenging issue in FSL. The weak feature representation leads to weak generalization and troubles in cross-domain. In this work, we propose a feature representation enhancement network (FREN) that enhances the feature representation through instance embedding and task adaptation. For instance embedding, global max pooling, and global average pooling are used together for adding more features, and Gaussian-like calibration is used for normalizing the feature distribution. For task adaptation, self-attention is adopted for task contextualization. Given the absence of publicly available data on tobacco, we created a tobacco leaf abnormality dataset (TLA), which includes 16 categories, two settings, and 1,430 images in total. In experiments, we use PlantVillage, which is the benchmark dataset for plant disease identification, to validate the superiority of FREN first. Subsequently, we use the proposed method and TLA to analyze and discuss the abnormality identification of tobacco. For the multi-symptom diseases that always have low accuracy, we propose a solution by dividing the samples into subcategories created by symptom. For the 10 categories of tomato in PlantVillage, the accuracy achieves 66.04% in 5-way, 1-shot tasks. For the two settings of the tobacco leaf abnormality dataset, the accuracies were achieved at 45.5% and 56.5%. By using the multisymptom solution, the best accuracy can be lifted to 60.7% in 16-way, 1-shot tasks and achieved at 81.8% in 16-way, 10-shot tasks. The results show that our method improves the performance greatly by enhancing feature representation, especially for tasks that contain categories with high similarity. The desensitization of data when crossing domains also validates that the FREN has a strong generalization ability.

DMNet: A Personalized Risk Assessment Framework for Elderly People With Type 2 Diabetes.

Type 2 diabetes is the most common chronic disease for the elderly people. This disease is difficult to be cured and causes continued medical expenses. The early and personalized risk assessment of type 2 diabetes is necessary. So far, various type 2 diabetes risk prediction methods have been proposed. However, these methods have three major issues: 1) not fully considering the importance of personal information and rating information of healthcare system, 2) not adopting the long-term temporal information, and 3) not comprehensively capturing the correlation between the diabetes risk factor categories. To address these issues, the personalized risk assessment framework for elderly people with type 2 diabetes is needed. However, it is very challenging due to two reasons, namely imbalanced label distribution and high-dimensional features. In this paper, we propose diabetes mellitus network framework (DMNet) for type 2 diabetes risk assessment of elderly people. Specifically, we propose tandem long short-term memory to extract the long-term temporal information of different diabetes risk categories. In addition, the tandem mechanism is used to capture the correlation between the diabetes risk factor categories. To balance the label distribution, we adopt the method of synthetic minority over-sampling technique with Tomek links. To form the better feature representations, we utilize entity embedding to solve the problem of high-dimensional features. To evaluate the performance of our proposed method, we conduct the experiments on a real-world dataset called Research on Early Life and Aging Trends and Effects. The experiment results show that DMNet outperforms the baseline methods in terms of six evaluation metrics (i.e., accuracy of 0.94, balanced accuracy of 0.94, precision of 0.95, F1-score of 0.95, recall of 0.95 and AUC of 0.94).

Few-Shot Learning for Plant-Disease Recognition in the Frequency Domain.

Few-shot learning (FSL) is suitable for plant-disease recognition due to the shortage of data. However, the limitations of feature representation and the demanding generalization requirements are still pressing issues that need to be addressed. The recent studies reveal that the frequency representation contains rich patterns for image understanding. Given that most existing studies based on image classification have been conducted in the spatial domain, we introduce frequency representation into the FSL paradigm for plant-disease recognition. A discrete cosine transform module is designed for converting RGB color images to the frequency domain, and a learning-based frequency selection method is proposed to select informative frequencies. As a post-processing of feature vectors, a Gaussian-like calibration module is proposed to improve the generalization by aligning a skewed distribution with a Gaussian-like distribution. The two modules can be independent components ported to other networks. Extensive experiments are carried out to explore the configurations of the two modules. Our results show that the performance is much better in the frequency domain than in the spatial domain, and the Gaussian-like calibrator further improves the performance. The disease identification of the same plant and the cross-domain problem, which are critical to bring FSL to agricultural industry, are the research directions in the future.

Few-shot learning approach with multi-scale feature fusion and attention for plant disease recognition.

Image-based deep learning method for plant disease diagnosing is promising but relies on large-scale dataset. Currently, the shortage of data has become an obstacle to leverage deep learning methods. Few-shot learning can generalize to new categories with the supports of few samples, which is very helpful for those plant disease categories where only few samples are available. However, two challenging problems are existing in few-shot learning: (1) the feature extracted from few shots is very limited; (2) generalizing to new categories, especially to another domain is very tough. In response to the two issues, we propose a network based on the Meta-Baseline few-shot learning method, and combine cascaded multi-scale features and channel attention. The network takes advantage of multi-scale features to rich the feature representation, uses channel attention as a compensation module efficiently to learn more from the significant channels of the fused features. Meanwhile, we propose a group of training strategies from data configuration perspective to match various generalization requirements. Through extensive experiments, it is verified that the combination of multi-scale feature fusion and channel attention can alleviate the problem of limited features caused by few shots. To imitate different generalization scenarios, we set different data settings and suggest the optimal training strategies for intra-domain case and cross-domain case, respectively. The effects of important factors in few-shot learning paradigm are analyzed. With the optimal configuration, the accuracy of 1-shot task and 5-shot task achieve at 61.24% and 77.43% respectively in the task targeting to single-plant, and achieve at 82.52% and 92.83% in the task targeting to multi-plants. Our results outperform the existing related works. It demonstrates that the few-shot learning is a feasible potential solution for plant disease recognition in the future application.

Edge-Based Transfer Learning for Classroom Occupancy Detection in a Smart Campus Context.

Studies and systems that are aimed at the identification of the presence of people within an indoor environment and the monitoring of their activities and flows have been receiving more attention in recent years, specifically since the beginning of the COVID-19 pandemic. This paper proposes an approach for people counting that is based on the use of cameras and Raspberry Pi platforms, together with an edge-based transfer learning framework that is enriched with specific image processing strategies, with the aim of this approach being adopted in different indoor environments without the need for tailored training phases. The system was deployed on a university campus, which was chosen as the case study. The proposed system was able to work in classrooms with different characteristics. This paper reports a proposed architecture that could make the system scalable and privacy compliant and the evaluation tests that were conducted in different types of classrooms, which demonstrate the feasibility of this approach. Overall, the system was able to count the number of people in classrooms with a maximum mean absolute error of 1.23.