Strawberry is a kind of fruit with high nutritional value, which has high economic effect. With the expansion of strawberry planting area year by year, strawberry diseases caused by unreasonable cultivation methods are also aggravated. In the process of strawberry disease control and management, grey mold and anthracnose are among the most serious diseases. Gray mold can infect under low temperature and humidity, and the invasion port is mostly in the plant wound and dead part, resulting in strawberry fruit decay. Anthrax can cause local spots on the leaves of strawberry plants. In severe cases, the whole plant will wither and die [2, 37].
A common method for controlling strawberry disease is based on laboratory tests, which includes Indicator plant leaflet grafting assay, electron microscopy assay and molecular biology polymerase chain assay (PCR) [55]. These scientific methods are more accurate than visual inspection, but most of them are inefficient, destructive and time-consuming, requiring precision instruments and rigorous operation. Therefore, accurate and non-destructive identification of grey mold and anthracnose is critical to strawberry production management [23, 61].
Hyperspectral imaging, a relatively new non-destructive detection technique, has been proved to have a wide application prospect in the detection of plant diseases. H Hyperspectral imaging technology combines imaging technology and spectral technology to obtain continuous and narrow band image data and spectral information of each pixel [19, 24]. In the field of agricultural products research, hyperspectrum can be used for qualitative analysis or quantitative detection. At present, the research reports of domestic and foreign scholars cover the analysis of external indicators such as surface defect damage, excreta pollution and the detection of internal indicators [5, 43, 50, 51, 56]. Several studies on the detection of strawberry diseases utilizing hyperspectral imaging were conducted. For instance, [38] applied identify and classify grapevines inoculated with the newly discovered DNA virus grapevine vein-clearing virus (GVCV) at the early asymptomatic stages utilizing hyperspectral imaging. The results showed that the automatic three-dimensional convolutional neu zral network (3d-cnn) feature extractor shows better learning effect than the two-dimensional convolutional neural network (2d-cnn) in learning features of hyperspectral data cubes under certain conditions. [8] obtained Hyperspectral Vegetation Index for leaf spot detection by identifying sensitive bands. The above results show that hyperspectral imaging technology is highly feasible for detecting the occurrence of peanut leaf spot. [11] proposed a non-destructive citrus HLB field detection method based on hyperspectral reflectance, among which the accuracy of SVM learner achieved 90.8% and 96% in the two groups of classification (healthy and symptomatic HLB leaves). The results showed that 13 characteristic bands extracted by the proposed method provided the best performance. [17] utilized the spectral features and texture features (TFs) of hyperspectral images to establish a SVM classifier and applied it to the recognition of yellow rust in Wheat Leaves. The study presented that the OA of the TFs scheme could reach 86.5%. [25] presented an early detection model based on machine learning technology of hyperspectral images, extracted the normalized difference texture index (ndtis) and vegetation index (VIS), and then used the partial least squares linear discriminant analysis method to detect and model powdery mildew with the combined optimal features (VIs and ndtis). The results showed that the discrimination model, visa and ndtis improve the ability of early recognition. The overall accuracy reached 82.35% and kappa coefficient reached 0.56, which has significantly improved compared with the early stage. [28] presented an extended collaborative representation (ECR)-based classification model, among which the diagnostic accuracy is higher than 94.7% and the average online diagnosis time is only 1-1.3ms, which can quickly and accurately diagnose Cucumber Leaf Diseases. The result indicated that the classification model based on ECR has strong applicability in cucumber leaf disease detection. [39] performed spectral angle mapping to segment the infected area from sound tissue, and monitored the pathogenetic process of the disease.
The above studies illustrate that hyperspectral imaging technology is feasible for detecting strawberry diseases. However, most of above diseases detection strategies only consider single feature such as spectrum, VIs and texture, which are not well adaptable to unstructured environments. The method of combining diverse features, such as spectral features and VI features, to identify strawberry diseases is less studied. Consequently, the main purpose of this work was to establish a robust methodology for early detection of strawberry leaves disease based upon hyperspectral imaging. The objectives are to: (1) Collect hyperspectral images of healthy leaves and infected leaves using a hyperspectral imaging system; (2) Extract spectral features and VIs from the preprocessed hyperspectral images; (3) Select the spectral fingerprint features and significant VIs features using CARS and ReliefF algorithm, respectively; (4) Combine the spectral fingerprint features and significant VIs features as inputs of diverse machine learning models for identification of strawberry leaves.