• Journal of Intelligence and Information Systems
• /
• v.25 no.1
• /
• pp.163-177
• /
• 2019

As smartphones are getting widely used, human activity recognition (HAR) tasks for recognizing personal activities of smartphone users with multimodal data have been actively studied recently. The research area is expanding from the recognition of the simple body movement of an individual user to the recognition of low-level behavior and high-level behavior. However, HAR tasks for recognizing interaction behavior with other people, such as whether the user is accompanying or communicating with someone else, have gotten less attention so far. And previous research for recognizing interaction behavior has usually depended on audio, Bluetooth, and Wi-Fi sensors, which are vulnerable to privacy issues and require much time to collect enough data. Whereas physical sensors including accelerometer, magnetic field and gyroscope sensors are less vulnerable to privacy issues and can collect a large amount of data within a short time. In this paper, a method for detecting accompanying status based on deep learning model by only using multimodal physical sensor data, such as an accelerometer, magnetic field and gyroscope, was proposed. The accompanying status was defined as a redefinition of a part of the user interaction behavior, including whether the user is accompanying with an acquaintance at a close distance and the user is actively communicating with the acquaintance. A framework based on convolutional neural networks (CNN) and long short-term memory (LSTM) recurrent networks for classifying accompanying and conversation was proposed. First, a data preprocessing method which consists of time synchronization of multimodal data from different physical sensors, data normalization and sequence data generation was introduced. We applied the nearest interpolation to synchronize the time of collected data from different sensors. Normalization was performed for each x, y, z axis value of the sensor data, and the sequence data was generated according to the sliding window method. Then, the sequence data became the input for CNN, where feature maps representing local dependencies of the original sequence are extracted. The CNN consisted of 3 convolutional layers and did not have a pooling layer to maintain the temporal information of the sequence data. Next, LSTM recurrent networks received the feature maps, learned long-term dependencies from them and extracted features. The LSTM recurrent networks consisted of two layers, each with 128 cells. Finally, the extracted features were used for classification by softmax classifier. The loss function of the model was cross entropy function and the weights of the model were randomly initialized on a normal distribution with an average of 0 and a standard deviation of 0.1. The model was trained using adaptive moment estimation (ADAM) optimization algorithm and the mini batch size was set to 128. We applied dropout to input values of the LSTM recurrent networks to prevent overfitting. The initial learning rate was set to 0.001, and it decreased exponentially by 0.99 at the end of each epoch training. An Android smartphone application was developed and released to collect data. We collected smartphone data for a total of 18 subjects. Using the data, the model classified accompanying and conversation by 98.74% and 98.83% accuracy each. Both the F1 score and accuracy of the model were higher than the F1 score and accuracy of the majority vote classifier, support vector machine, and deep recurrent neural network. In the future research, we will focus on more rigorous multimodal sensor data synchronization methods that minimize the time stamp differences. In addition, we will further study transfer learning method that enables transfer of trained models tailored to the training data to the evaluation data that follows a different distribution. It is expected that a model capable of exhibiting robust recognition performance against changes in data that is not considered in the model learning stage will be obtained.

• KOREAN JOURNAL OF CROP SCIENCE
• /
• v.11
• /
• pp.81-97
• /
• 1972

Experiments and investigations were done basically and practically for the purpose of labor saving in paddy rice cultivation especially on Homizil i.e. hoeing and herbicide, 1969. 8 concrete tanks were established on the open base of Keon Kuk University for comparison of percolation, dissolved oxygen and yield test of rice in the paddy plot of tank. The dimension of the bottom of each tank is square meter. Each of the 4 of the 8 tanks is 21cm in height and each of the remaining 4 tanks is 36cm. Each tank has a system that comprises 2 sets of tubes, each of which has 20 holes of 5mm in diameter scattered every side and is covered with nylon cloth taking water in the tank. One set consists of 4 P.V.C tubes. The first set is situated 8cm below the top of the tank and the second set is located at bottom layer inside the tank. The 4 tubes of each set are combined together and led to the glass tube which protects from inside to outside. And this inside-outside glass tube is connected to the small rubber tube. Also a glass tube is set 4cm below the top of the tank. Paddy loam was filled on sand in each of the tanks in the soil depth of either 15cm or 30cm. The depth of sand was 5cm in the soil depth of 15cm and 10cm in the soil depth of 30cm. (Fig. 1, 2 and 3). The paddy rice was grown in the tank. The percolation of water, the dissolved oxygen and the yield of rice were observed in the tank. And the dissolved oxygen was detected by Winkler method. A sandy paddy field of heavy percolation was selected at the field of the National Agricultural Material Inspection Center in Seoul. It was divided into 9 plots. These plots were given 3 treatments: (A) not hoeing, (B) hoeing one time and (C) hoeing two times. These treatments were replicated 3 times along the latin square design. The paddy rice was grown and sprayed with Stam F-34 in the all plots for the purpose of killing weeds before hoeing. The two types of paddy of field i.e. one for normal percolation and the other for ill drainage were selected at Iri Crop Experiment Station, Jeonla-Bukdo. Each field was divided into 24 plots for 8 treatments. They are: (A) not hoeing; (B) hoeing one time; (C) hoeing two times; (D) not hoeing but treating with herbicide, Pamcon; (E) hoeing one time and weeding two times also treating with herbicide, Pamcon; (F) hoeing two times and weeding one time a], o treating with herbicide, Pamcon; (G) hoeing two times and weeding two times also treating with herbicide, Pamcon, ; (H) usual manner. The labor hours and expenses needed for weeding in the paddy by hoeing were investigated in a farmer at Suwon and the price of herbicide and the yield of rice were taken out at Iri, Jeonla-Bukdo. The results obtained from the above experiments and investigations are as follows: 1. The relationship between percolation and dissolved oxygen shows that a very small amount of oxygen is detected in the soil water under 2cm below surface of earth in the paddy even when percolation is over 4.0cm per 24 hours (Tab. 1). 2. The relationship between percolation and yield of rice shows that the yield of rice increases in the percolation of 0cm and 1.5cm per 24 hours and decreases in the percolation of 2.5cm and 3.4cm in the plot of the 15cm ploughing depth and increases in the percolation of 1.4cm and 3.0cm and decreases in the percolation of 0cm and 4.0cm in the plot of 30cm ploughing depth (Tab. 1 and Fig. 5). 3. The yield of paddy weeded with Stam F-34 in the sandy field of heavy percolation in Seoul was 3.02 tons in the plot of not hoeing, 2.99 tons in hoeing one time and 3.05 tons in hoeing two times per hectare (Tab. 5). 4.1). 4. 1) The yield of rice per 10 ares in the field of normal percolation at Iri was 338kg in not hoeing, 379kg in hoeing one time, 383kg in hoeing two times, 413kg in spraying herbicide, Pamcon, and not hoeing, 433kg in spraying herbicide, Pamcon, and hoeing one time and weeding two times, 399kg in spraying herbicide, Pamcon, and hoeing two times and weeding one time, 420kg in spraying herbicide, Pamcon, and hoeing two times and weeding two times and 418kg in usual manner (Tab. 6-1). 2) The yield of rice per 10 ares in the field of ill drainage at Iri was 323kg in not hoeing, 363kg in hoeing one time, 342kg in hoeing two times, 388kg in spraying herbicide, Pamcon, and not hoeing, 425kg in spraying herbicide, Pamcon, and hoeing one time and weeding two times, 427kg in spraying herbicide, Pamcon, and hoeing two times and weeding one time, 449kg in spraying herbicide, Pamcon, and hoeing two times and weeding two times and 412kg in usual manner (Tab. 6-2). 5. 1) The labor hours for weeding by hoeing was 37.1 hours but 53.5 hours if hours for meal, smoking and so on are included, and the expenses including labor cost needed for weeding by hoeing in the paddy rice was 2, 346 Won per 10 ares at Suwon (Tab. 7). 2) The labor hours for weeding by spraying herbicide with hand sprayer in the paddy rice was about 5 hours per 10 ares at Suwon and the expenses for weeding by spraying herbicide in the paddy rice was 750 Won but 1130 Won if the loss by decrement of rice in the paddy field of ill drainage per 10 ares is calculated in estimation at Iri (Tab. 8). From these observations and investigations it is known that using of some kinds of herbicides Saves labor and expenses of weeding, almost without giving damages to the rice itself, in the field of normal or heavy percolation comparing usual manner of hoeing.