The Recognition Of Stored Grain Pests Based On Sparse Representation

Food security is great importance for economic development and social stability. Thevigorous development of science and technology make the food production has a very largeincrease. The reserves of grain has also increasing. Improper food storage can directly causehuge losses for the state and the people. Grain pest hazards is one important reason of grainstored loss. Only accurate detection and correct identification for grain insects, can preventionand treatment have a purpose. In the classification and recognition of grain insects, theredundancy of feature will increase the complexity and computational of algorithms.Reconstruction algorithms have the contradictions of time consuming and classificationsuccess rate in the practical application. The classification method based on sparsecharacterization, which derived from the theory of compressed sensing, based on a reasonablefeature extraction can better achieve grain insect classification. This paper aiming to reducefeatures redundancy and optimization algorithm, has completed the following tasks in theframework of compressed sensing:1About the feature extraction, make image pre-processing of Pirates of Otani, lessergrain borer, black fungus and other9kinds of common granary insect pests, Gabor energyfeatures are extracted. Simulation results show that, Gabor energy feature is a better choosefor grain pests classification.2About the optimization of features, the overall situation, different scales, differentdirections three aspects are considered. Local Gabor features are extraction according to theamount of energy. Principal component analysis is used to on the extracted features for furtherdimension reduction and optimization. Ultimately, short time-consuming and highclassification success rate features are selected, provides the basis of the dictionary structure.3About the reconstruction algorithm, reconstruction model of the convex relaxationalgorithm (L1) and the greedy algorithm are description, simulation under different programof features. By the analysis of theL1algorithm outperforms ROMP algorithm and timeconsuming of the SP algorithm is less thanL1algorithm, K=6, K=8are selected as thecombination algorithms sparsity value.4About the optimization algorithm, a algorithm based onL1and ROMP algorithm, aalgorithm based onL1and SP algorithm are proposed, the corresponding theoretical modelare given. Then simulation under different program of features. Simulation results show that, the new program and the proposed algorithm has a better recognition results, which providesa theoretical basis for the sparse characterization in grain insects recognition.