Assessing crop damage from dicamba on non-dicamba-tolerant soybean by hyperspectral imaging through machine learning

Authors: ZHANG, JINGCHENG - Hangzhou Dianzi University; Huang, Yanbo; Reddy, Krishna; WANG, BIN - Hangzhou Dianzi University

Submitted to: Pest Management Science
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 4/22/2019
Publication Date: 11/6/2019
Citation: Zhang, J., Huang, Y., Reddy, K.N., Wang, B. 2019. Assessing crop damage from dicamba on non-dicamba-tolerant soybean by hyperspectral imaging through machine learning. Pest Management Science. 75:3260-3272.

Interpretive Summary: Increased use of dicamba has potential to cause injury to susceptible crops from off-target drift. A rapid, cost-effective method is needed for assessment of crop injury from dicamba on large scale farms. Scientists of Hangzhou Danzi University, China and USDA ARS Crop Production Systems Research Unit at Stoneville, Mississippi have collaboratively developed a hyperspectral imaging method through machine learning algorithms to identify sensitive bands to the dicamba spray rates and categorize the rates into the recoverable and not recoverable groups. The results of this research give high accuracy in differentiation of health to damaged plants at different severity, which can be used to provide useful information for weed management.

Technical Abstract: BACKGROUND: Dicamba (3, 6-dichloro-2-methoxybenzoic acid) effectively controls several broadleaf weeds. The off-target drift of dicamba spray or vapor drift can cause severe injury to susceptible crops including non-dicamba-tolerant crops. The assessment of the crop damage from dicamba drift is critical for effective weed management. In a field experiment, advanced hyperspectral imaging (HSI) technique was used to study the spectral response of soybean plants to different dicamba rates, and developed appropriate spectral features and models for assessing the crop damage from dicamba. RESULTS: An experiment with 6 different dicamba rates, an ordinal spectral variation pattern was measured at both 1 week after treatment (WAT) and 3 WAT. The soybean receiving a dicamba rate = 0.2X exhibited an unrecoverable damage. Two recoverability spectral indices (HDRI, HDNI) were developed based on three optimal wavebands. Based on Jeffries-Matusita distance metric, Spearman correlation analysis and independent t-test for sensitivity to dicamba spray rates, a number of wavebands were identified, derivative and continuous removal features were extracted, and vegetation indices were calculated. The models for quantifying dicamba spray levels were established using machine learning algorithms of Naive Bayes, Random Forest and Support Vector Machine. CONCLUSIONS: The spectral response of soybean injury caused by dicamba sprays can be clearly captured by HSI technique. The developed recoverability spectral indices were able to accurately differentiate the recoverable and unrecoverable damages, with an overall accuracy (OA) higher than 90%. The optimal spectral feature sets were identified for characterizing dicamba spray rates under recoverable and unrecoverable situations. The spectral features plus plant height can yield relatively high accuracy under recoverable situation (OA=94%). These results can be of practical importance in weed management.