Nowadays, avocado has strong demand around the world due to its nutritional properties and because it is all year supplied from different parts of the world, being Peru one of the main providers. However, nutrient deficiencies and plague attacks during cultivation stages represent a major difficulty for farmers since early identification of these states (i.e. deficiencies and plagues) is a time-consuming activity that requires trained evaluators to do so. In this paper, an automatic method for identification of avocado leaf state is proposed. This method uses k-means, in a s-v space at superpixel level, to segment leaf from uniform background from images captured in-field in semi-controlled conditions and a shallow neural network to classify composed histograms from segmented leaves into 4 states: Healthy, Fe deficiency, Mg deficiency and red spider plague. The proposed method separates leaf from background with an average F-score of 0.98 and classifies leaf condition with an overall accuracy of 96.8%.