Song Yong, Chen Bing, Wang Qiong, Wang Jing, Zhao Jing, Sun Lexin, Chen Zijie, Han Huanyong, Wang Fangyong, Fu Jihai. Estimation of yield loss in diseased cotton fields using UAV multi-spectral images[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2022, 38(6): 175-183. DOI: 10.11975/j.issn.1002-6819.2022.06.020
    Citation: Song Yong, Chen Bing, Wang Qiong, Wang Jing, Zhao Jing, Sun Lexin, Chen Zijie, Han Huanyong, Wang Fangyong, Fu Jihai. Estimation of yield loss in diseased cotton fields using UAV multi-spectral images[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2022, 38(6): 175-183. DOI: 10.11975/j.issn.1002-6819.2022.06.020

    Estimation of yield loss in diseased cotton fields using UAV multi-spectral images

    • Abstract: Verticillium wilt has been one of the most common high-risk diseases of cotton in recent years. A huge threat can be posed to the stable production of cotton fields in Xinjiang, one of the most important commodity cotton production bases in China. However, the annual area of verticillium wilt disease in cotton fields is ever increasing and resulting in serious yield losses, due to the long-term influence of planting modes, geographical, and climate characteristics. Most traditional measurements for the yield loss present time-consuming, labor-intensive, large errors, one-sidedness. It is a high demand to accurately estimate the yield loss caused by the diseases. Fortunately, the Unmanned Aerial Vehicle (UAV) multi-spectral data can be combined with the ground survey data for the comprehensive characteristics of high accuracy. In this study, a systematic investigation was conducted to estimate the yield loss in the diseased cotton fields using the UVA multi-spectral images. An experiment was carried out in the cotton verticillium wilt disease nursery (44°31′N, 85°98′E) of Xinjiang Academy of Agricultural Sciences, Shihezi Reclamation District, Xinjiang in China in 2020. The gray desert soil was collected in the study area, with an organic matter content of 21.30 g/kg, a soil layer pH of 7.82, and electrical conductivity (EC) of 0.48 mS/cm. Taking the test variety as Xinluzao 8, the specific procedure was set: the on-demand sowing was adopted on the film, the drip irrigation under the film, (66 +10) cm wide and narrow row design, while the cotton plants were manually capped on July 5, and sprayed by drones on September 6 and September 13 Defoliant, and finally harvested by the cotton picker on October 13th. The ground monitoring points were arranged in a grid format (66 in total) from the beginning of the disease, in order to collect the disease severity data. The output was then measured in the later period. The monitoring point was taken as the center during the production measurement (cotton harvest period), and the area of the monitoring point was expanded to 6.67m2 for the yield measurement. The UAV data acquisition was consistent with the acquisition time of ground monitoring points. The correlation coefficient method was used to screen the optimal vegetation index for the diseased cotton plants, according to the UAV multi-spectral images of diseased cotton fields and ground yield loss data. The gray value standard deviation method was also used to screen the optimal band combination. As such, the comprehensive cotton field image was established, including the best band combination and DVI comprehensive image. The support vector machine radial basis kernel function classification was used to analyze the spatial distribution of the original image and the comprehensive image of the diseased cotton field, further estimate the yield loss. The results show that the best vegetation index and the best band combination were DVI (correlation coefficient |r|=0.86) and B3-B5-B8 (optimum index factor was 153.44) for the UAV multi-spectral image identification of diseased cotton fields. The comprehensive image accurately identified the spatial distribution of diseased cotton fields, where the accuracy of the bolling period was the highest (overall accuracy was 96.64%, Kappa coefficient was 95.61%). In the different disease severity (health b0, slight b1, moderate b2, serious b3, and critical b4) corresponding to the cotton field, the area ratios were 7.81%, 23.78%, 29.20%, 13.92%, and 17.43%, compared with the original image. Consequently, the comprehensive image performed the best to estimate the yield loss of diseased cotton fields. The different disease severity (b0, b1, b2, b3, and b4) corresponded to 0.00%, 22.80%, 31.32%, 49.02%, and 76.33%, respectively. It was estimated that the loss of seed cotton was 4260.01 kg, and the loss rate was 49.16%, and the loss of lint cotton was 2267.18 kg, and the loss rate was 54.51%. Compared with the estimated yield loss of cotton fields under disease stress, the actual loss rate of seed cotton in cotton fields was 6.28% higher, and the loss rate of lint cotton was 4.48% higher. The estimated value of disease-stressed cotton field yield is not significantly different from the actual harvest value of cotton field, which can accurately estimate the yield loss of diseased cotton field. Therefore, it is possible to accurately estimate the yield loss of diseased cotton fields using the integrated UAV image. The finding can provide a strong reference for the estimation of yield loss caused by similar diseases.
    • loading

    Catalog

      /

      DownLoad:  Full-Size Img  PowerPoint
      Return
      Return