Deep Convolutional Neural Network for Rice Density Prescription Map at Ripening Stage Using Unmanned Aerial Vehicle-Based Remotely Sensed Images

Abstract

In this paper, UAV (unmanned aerial vehicle, DJI Phantom4RTK) and YOLOv4 (You Only Look Once) target detection deep neural network methods were employed to collected mature rice images and detect rice ears to produce a rice density prescription map. The YOLOv4 model was used for rice ear quick detection of rice images captured by a UAV. The Kriging interpolation algorithm was used in ArcGIS to make rice density prescription maps. Mature rice images collected by a UAV were marked manually and used to build the training and testing datasets. The resolution of the images was 300 × 300 pixels. The batch size was 2, and the initial learning rate was 0.01, and the mean average precision (mAP) of the best trained model was 98.84%. Exceptionally, the network ability to detect rice in different health states was also studied with a mAP of 95.42% in the no infection rice images set, 98.84% in the mild infection rice images set, 94.35% in the moderate infection rice images set, and 93.36% in the severe infection rice images set. According to the severity of rice sheath blight, which can cause rice leaves to wither and turn yellow, the blighted grain percentage increased and the thousand-grain weight decreased, the rice images were divided into these four infection levels. The ability of the network model (R2 = 0.844) was compared with traditional image processing segmentation methods (R2 = 0.396) based on color and morphology features and machine learning image segmentation method (Support Vector Machine, SVM R2 = 0.0817, and K-means R2 = 0.1949) for rice ear counting. The results highlight that the CNN has excellent robustness, and can generate a wide range of rice density prescription maps.

Comparison of Deep Learning Methods for Detecting and Counting Sorghum Heads in UAV Imagery

“…The main influencing factor in terms of UAV images was the spatial resolution of the images [4], which was an important indicator affecting the accuracy of sorghum head detection, and was mainly affected by the UAV flight altitude [8], which in this study was 20 m and the spatial resolution of the image was 1.1 cm. Compared to existing studies on crop spike detection and counting that commonly used millimeter-level (generally less than 5 mm) spatial resolution images [4,6,10,18,20,22,71], the sorghum head targets in this study were smaller and their boundaries were more blurred in the images used in this study at the centimeter level. The lower spatial resolution directly affected the learning performance of the DL methods [72,73], which undoubtedly hindered sorghum head detection and counting from the UAV images.…”

“…Compared to EfficientDet, which required higher clarity of sorghum heads, and the SSD network, which predicted targets for each feature layer individually, the YOLOv4 algorithm had better sorghum head detection results. The YOLO series of algorithms has been used for detecting corn plant seedlings [19], rice ears [10], cotton seedlings [63], cherry fruit [64], apples and apple flowers [49,51], and greenhouses [65], all of which have more extensive applications and better detection results. However, there are more convolutional layers stacked on each other in the CSPDarknet53 backbone network of YOLOv4, so that YOLOv4 had a larger number of parameters and floating-point computation, and its number of parameters was the largest among the three methods.…”

“…The main indicators that affect the detection accuracy of sorghum heads themselves are planting density and growth period [3,4,74]. (1) In terms of planting density, it has been generally accepted that as planting density increases, crops overlap and shade each other, which, in severe cases, causes some duplicate detection, a large number of missed detections, underestimation of crop counts, and a reduction in the performance of DL methods [5,8,10,18]. However, the opposite conclusion was drawn in this study and it was found that the detection accuracy tended to be worse for moderate and sparse conditions than for dense conditions (Figure 6).…”

“…With the rapid development of remote sensing platforms and sensor technology, imagery with high spatial and temporal resolutions can be obtained to detect and quantify crop ears in a low-cost and easy-to-use manner by unmanned aerial vehicles (UAVs) [10][11][12][13], which are flexible for image acquisition and unaffected by cloud and terrain conditions. UAVs can usually be equipped with a variety of sensors, such as RGB, multispectral, hyperspectral, thermal, and Light Detection and Ranging (LiDAR) [14].…”

Rice Crop Counting Using Aerial Imagery and GIS for the Assessment of Soil Health to Increase Crop Yield

“…In [ 35 ], with the growing availability of RGB data with extremely high spatial resolution, this study demonstrated the efficiency of a deep convolutional neural network technique for producing rice density prescription maps using UAV-based imagery. For counting and geolocation, their solution in [ 36 ] performs noticeably better than competing object detection techniques.…”