This paper proposes a system for the forecasting and automated inspection of rice Bakanae disease (RBD) infection rates via drone imagery. The proposed system synthesizes camera calibrations and area calculations in the optimal data domain to detect infected bunches and classify infected rice culm numbers. Optimal heights and angles for identification were examined via linear discriminant analysis and gradient magnitude by targeting the morphological features of RBD in drone imagery. Camera calibration and area calculation enabled distortion correction and simultaneous calculation of image area using a perspective transform matrix. For infection detection, a two-step configuration was used to recognize the infected culms through deep learning classifiers. The YOLOv3 and RestNETV2 101 models were used for detection of infected bunches and classification of the infected culm numbers, respectively. Accordingly, 3 m drone height and 0° angle to the ground were found to be optimal, yielding an infected bunches detection rate with a mean average precision of 90.49. The classification of number of infected culms in the infected bunch matched with an 80.36% accuracy. The RBD detection system that we propose can be used to minimize confusion and inefficiency during rice field inspection.

Recently, the analysis research of crop's growth condition is done with the use of hyperspectral image. However, there are many factors such as physical factors and complexity of data make the hyperspectral image analysis difficult. This study presents the classification method of crop's leaf growth condition using hyperspectral image(HSI) and Deep Neural Network(DNN). Major information of plants is acquired through hyperspectral image, and the preprocessing is followed for the information to be used for DNN learning. The preprocessing is used by cutting the data in small patch size and rotating it for the models to be operated effectively. In the experiment, paprika leaves are divided into four types of leaves and backgrounds such as normal and damaged by harmful insects, and the result of the experiment showed 90.9% of accuracy. The presented method has advantages that the data generation method does not affect DNN and can classify various growth conditions that are difficult in the existing RGB image.

Keunho Park, Hee-Shin Lee, and Joonwhoan Lee. IJFIS 2015;15:102-10. https://doi.org/10.5391/IJFIS.2015.15.2.102

Rice lodging, a common physiological issue that occurs during rice growth and development, is a major factor contributing to a decline in rice production. Current techniques for the extraction of rice lodging are subjective and require considerable time and labor. In this paper, we propose a fully automated method in an end-to-end format to objectively calculate the rice-lodging rate based on remote sensing data captured by a drone under field conditions. An image post-processing method was applied to enhance the semantic-segmentation results of an operable lightweight model on an embedded board. The area of interest within the parcel was preserved based on these results, and the lodging occurrence rate was calculated in a fully automated manner without external intervention. Five models were compared based on the U-Net and lite-reduced atrous spatial pyramid pooling (LR-ASPP) models with MobileNet versions 1–3 as the backbones. The final model, MobileNetV1_U-Net, performed the best with an RMSE of 11.75 and R² of 0.875, and MobileNetV3_LR-ASPP (small) achieved the shortest processing time of 4.9844 s. This study provides an effective method for monitoring large-scale rice lodging, accurate extraction of areas of interest, and calculating lodging occurrence rates.

논문은 실시간 돌발 상황 정보 인식이 어려운 터널 내에서 CCTV를 통해 실시간으로 돌발 상황을 인식할 수 있는 시스템을 개발하였다. 돌발 상황은 차량의 역주행 및 정지, 사람의 보행, 화재 및 낙하물의 총 5가지 상황으로 정의하였다. 돌발 상황을 효율적으로 인식하기 위해 화재와 낙하물은 배경제거 알고리즘을 활용하고, 차량 및 사람은 YOLOv3를 기반으로 한 Deep-sort 알고리즘을 활용하는 하이브리드 방식의 영상인식 기술을 사용하였다. 각 돌발 상황에 대한 총 175번의 상황을 부여하여 실험을 진행하였고, 이 중 160번의 돌발상황을 인식하여 91%의 정확도를 보였다.