A deterministic technique for identifying dicotyledons in images

•Recognition, in a real uncontrolled environment, of climb weeds in sugarcane plantations.•Use of the Haar-like method to extract features that determine the difference between monocotyledonous and dicotyledonous.•Parameter Adjustment Using Fuzzy Logic.•Proposed methodology allowed recognition regardless of rotation and leaf size.•Low algorithmic complexity, suitable for restricted systems with real-time requirements. The identification of plants is often based on leaf recognition. Ipomoea spp., a dicotyledon weed present in sugarcane plantations, has unique vesiculated venation patterns that can be used in the recognition process. The uncontrolled plantation environment imposes challenges to leaf-based plant identification, such as overlap, light intensity, and occlusion. This work proposes a method for accurate and fast identification of leaves using Haar-like features, Fuzzy Logic, and Connected Components to differentiate monocotyledons and dicotyledons. Fuzzy Logic is used to define the template size for Haar-like features, combined with Integral Image concept to reduce processing time by lowering the arithmetic operation count. Our proposal was able to differentiate the target dicotyledonous leaves in an uncontrolled field image with more than one dicotyledon leaf. The obtained accuracy is acceptable regarding the current literature and the processing time was reduced.