A rapid and precise large-scale agricultural disaster survey is a basis for agricultural disaster relief and insurance but is labor-intensive and time-consuming. This study applies Unmanned Aerial Vehicles (UAVs) images through deep-learning image processing to estimate the rice lodging in paddies over a large area. This study establishes an image semantic segmentation model employing two neural network architectures, FCN-AlexNet, and SegNet, whose effects are explored in the interpretation of various object sizes and computation efficiency. Commercial UAVs imaging rice paddies in high-resolution visible images are used to calculate three vegetation indicators to improve the applicability of visible images. The proposed model was trained and tested on a set of UAV images in 2017 and was validated on a set of UAV images in 2019. For the identification of rice lodging on the 2017 UAV images, the F1-score reaches 0.80 and 0.79 for FCN-AlexNet and SegNet, respectively. The F1-score of FCN-AlexNet using RGB + ExGR combination also reaches 0.78 in the 2019 images for validation. The proposed model adopting semantic segmentation networks is proven to have better efficiency, approximately 10 to 15 times faster, and a lower misinterpretation rate than that of the maximum likelihood method.

Rice lodging identification relies on manual in situ assessment and often leads to a compensation dispute in agricultural disaster assessment. Therefore, this study proposes a comprehensive and efficient classification technique for agricultural lands that entails using unmanned aerial vehicle (UAV) imagery. In addition to spectral information, digital surface model (DSM) and texture information of the images was obtained through image-based modeling and texture analysis. Moreover, single feature probability (SFP) values were computed to evaluate the contribution of spectral and spatial hybrid image information to classification accuracy. The SFP results revealed that texture information was beneficial for the classification of rice and water, DSM information was valuable for lodging and tree classification, and the combination of texture and DSM information was helpful in distinguishing between artificial surface and bare land. Furthermore, a decision tree classification model incorporating SFP values yielded optimal results, with an accuracy of 96.17% and a Kappa value of 0.941, compared with that of a maximum likelihood classification model (90.76%). The rice lodging ratio in paddies at the study site was successfully identified, with three paddies being eligible for disaster relief. The study demonstrated that the proposed spatial and spectral hybrid image classification technology is a promising tool for rice lodging assessment.

Despite the existence of long term remotely sensed datasets, change detection methods are limited and often remain an obstacle to the effective use of time series approaches in remote sensing applications to Land Change Science. This paper establishes some simple statistical tests to be applied to NDVI-derived time series of remotely sensed data products. Specifically, the methods determine the statistical significance of three separate metrics of the persistence of vegetation cover or changes within a landscape by comparison to various forms of “benchmarks”; directional persistence (changes in sign relative to some fixed reference value), relative directional persistence (changes in sign relative to the preceding value), and massive persistence (changes in magnitude relative to the preceding value). Null hypotheses are developed on the basis of serially independent, normally distributed random variables. Critical values are established theoretically through consideration of the numeric properties of those variables, application of extensive Monte Carlo simulations, and parallels to random walk processes. Monthly pixel-level NDVI values for the state of Florida are analyzed over 25 years, illustrating the techniques’ abilities to identify areas and/or times of significant change, and facilitate a more detailed understanding of this landscape. The potential power and utility of such techniques is diverse within the area of remote sensing studies and Land Change Science, especially in the context of global change.

The scattered pixel problem in hyperspectral images caused by atmospheric noises and incomplete classification can lead to unsatisfactory classification; this problem remains to be solved. This letter reports the application of minimum noise fractions (MNFs) combined with fast and adaptive bidimensional empirical mode decomposition (FABEMD) as a two-step process to improve the classification accuracy of airborne visible-infrared imaging spectrometer hyperspectral image of the Indian Pine data set. With dimensional reduction by using MNF, FABEMD, considered as a low-pass filter, decomposes a hyperspectral image into several bidimensional intrinsic mode functions (BIMFs) and a residue image. The first four BIMFs are removed and the remainder BIMFs are integrated to reconstruct informative images that are subsequently classified through a support vector machine classifier (SVM). The classification results show that the proposed approach can effectively eliminate noise effects and can obtain higher accuracy than does traditional MNF SVM.