Application of Extreme Learning Machine to Visual Diagnosis of Rapeseed Nutrient Deficiency

Abstract

The nondestructive and accurate diagnosis of nutrient deficiency is the key to the adoption of appropriate remedial measures in agriculture. This paper proposes an intelligent visual diagnosis technique for the diagnosis of nutrient deficiency in rapeseed based on the leaf features. To this end, the experimental image library of four types of deficiencies was established, including normal, nitrogen deficiency, phosphorus deficiency and kalium deficiency. First, through employing the GrabCut algorithms, an image with remarkable features was divided into the foreground and background images. Then, the foreground one was employed to extract color, texture and shape features using the average grayscale and color moments in R/G/B and H/I channels, grayscale co-occurrence matrix and wavelet moments, respectively. Second, the initial features were normalized and filtered based on the gain for recognition rate so as to reduce their dimensions for improving the speed and accuracy of the diagnosis. Finally, the core features were imported into the extreme learning machine and the diagnosis of the input rapeseed leaf image could be accomplished. The experimental results showed that the proposed method could accurately identify the common nutrient deficiency, which sets a good example for the diagnosis technique of nutrient deficiency based on image analysis.