Locusts are agricultural pests found in many parts of the world. Developing efficient and accurate locust information acquisition techniques helps in understanding the relation between locust distribution density and structural changes in locust communities. It also helps in understanding the hydrothermal and vegetation growth conditions that affect locusts in their habitats in various parts of the world as well as in providing rapid and accurate warnings on locust plague outbreak. This study is a preliminary attempt to explore whether the batch normalization-based convolutional neural network (CNN) model can be applied used to perform automatic classification of East Asian migratory locust (AM locust), Oxya chinensis (rice locusts), and cotton locusts. In this paper, we present a way of applying the CNN technique to identify species and instars of locusts using the proposed ResNet-Locust-BN model. This model is based on the ResNet architecture and involves introduction of a BatchNorm function before each convolution layer to improve the network’s stability, convergence speed, and classification accuracy. Subsequently, locust image data collected in the field were used as input to train the model. By performing comparison experiments of the activation function, initial learning rate, and batch size, we selected ReLU as the preferred activation function. The initial learning rate and batch size were set to 0.1 and 32, respectively. Experiments performed to evaluate the accuracy of the proposed ResNet-Locust-BN model show that the model can effectively distinguish AM locust from rice locusts (93.60% accuracy) and cotton locusts (97.80% accuracy). The model also performed well in identifying the growth status information of AM locusts (third-instar (77.20% accuracy), fifth-instar (88.40% accuracy), and adult (93.80% accuracy)) with an overall accuracy of 90.16%. This is higher than the accuracy scores obtained by using other typical models: AlexNet (73.68%), GoogLeNet (69.12%), ResNet 18 (67.60%), ResNet 50 (80.84%), and VggNet (81.70%). Further, the model has good robustness and fast convergence rate.

In recent years, remote sensing (RS) research on crop growth status monitoring has gradually turned from static spectrum information retrieval in large-scale to meso-scale or micro-scale, timely multi-source data cooperative analysis; this change has presented higher requirements for RS data acquisition and analysis efficiency. How to implement rapid and stable massive RS data extraction and analysis becomes a serious problem. This paper reports on a Raster Dataset Clean & Reconstitution Multi-Grid (RDCRMG) architecture for remote sensing monitoring of vegetation dryness in which different types of raster datasets have been partitioned, organized and systematically applied. First, raster images have been subdivided into several independent blocks and distributed for storage in different data nodes by using the multi-grid as a consistent partition unit. Second, the “no metadata model” ideology has been referenced so that targets raster data can be speedily extracted by directly calculating the data storage path without retrieving metadata records; third, grids that cover the query range can be easily assessed. This assessment allows the query task to be easily split into several sub-tasks and executed in parallel by grouping these grids. Our RDCRMG-based change detection of the spectral reflectance information test and the data extraction efficiency comparative test shows that the RDCRMG is reliable for vegetation dryness monitoring with a slight reflectance information distortion and consistent percentage histograms. Furthermore, the RDCGMG-based data extraction in parallel circumstances has the advantages of high efficiency and excellent stability compared to that of the RDCGMG-based data extraction in serial circumstances and traditional data extraction. At last, an RDCRMG-based vegetation dryness monitoring platform (VDMP) has been constructed to apply RS data inversion in vegetation dryness monitoring. Through actual applications, the RDCRMG architecture is proven to be appropriate for timely vegetation dryness RS automatic monitoring with better performance, more reliability and higher extensibility. Our future works will focus on integrating more kinds of continuously updated RS data into the RDCRMG-based VDMP and integrating more multi-source datasets based collaborative analysis models for agricultural monitoring.

Nowadays, GF-1 (GF is the acronym for GaoFen which means high-resolution in Chinese) remote sensing images are widely utilized in agriculture because of their high spatio-temporal resolution and free availability. However, due to the transferrable rationale of optical satellites, the GF-1 remote sensing images are inevitably impacted by clouds, which leads to a lack of ground object’s information of crop areas and adds noises to research datasets. Therefore, it is crucial to efficiently detect the cloud pixel of GF-1 imagery of crop areas with powerful performance both in time consumption and accuracy when it comes to large-scale agricultural processing and application. To solve the above problems, this paper proposed a cloud detection approach based on hybrid multispectral features (HMF) with dynamic thresholds. This approach combined three spectral features, namely the Normalized Difference Vegetation Index (NDVI), WHITENESS and the Haze-Optimized Transformation (HOT), to detect the cloud pixels, which can take advantage of the hybrid Multispectral Features. Meanwhile, in order to meet the variety of the threshold values in different seasons, a dynamic threshold adjustment method was adopted, which builds a relationship between the features and a solar altitude angle to acquire a group of specific thresholds for an image. With the test of GF-1 remote sensing datasets and comparative trials with Random Forest (RF), the results show that the method proposed in this paper not only has high accuracy, but also has advantages in terms of time consumption. The average accuracy of cloud detection can reach 90.8% and time consumption for each GF-1 imagery can reach to 5 min, which has been reduced by 83.27% compared with RF method. Therefore, the approach presented in this work could serve as a reference for those who are interested in the cloud detection of remote sensing images.