Fusion of Deep Features for Yellow Rust Severity Estimation in Wheat Leaves

Wheat is one of the oldest cultivated plants that is considered as the most important staples for many nations. For this reason, enhancing both the quality and the quantity of the wheat corps is a state goal in many national economies. This is achieved through early diagnosis of wheat leaf diseases which is closely linked to food safety, environmental sustainability and economic stability. Specifically, the estimation of the disease severity helps to adjust the use of pesticides and makes a better disease monitoring. The Yellow Rust is the most widespread wheat leaf infection that leads to substantial crop losses. In this work, a severity estimation system in which we characterize the severity level by fusing deep features derived from two different CNN models. In a first step, VGG16 and MobileNetV2 are trained to generate images deep features. Then, generic fusion rules are employed to combine features which are subsequently trained by a SVM classifier. Experiments are conducted on Yellow-Rust-19 dataset, which contains 15000 wheat leaf images. Experimental results reveal that fusion rules give an improvement around 0.4% in total accuracy over the scores derived from individual CNN.