The application of remote sensing technique to identifying and classifying the quaternary sediments in south fringe of NAN XIANG basin
Enhancement of Remote Sensing information of quaternary strata and extraction of the feature information
The main signs of interpretation on remote sensing image for quaternary strata are the difference of grey, texture landforms and the river systems as well as the frequency features. The information of Quaternary strata in the original image which was influenced by various factors during image formation is weak and boundaries between the types are indistinct. In order to extract the information, the image must be enhanced. (a) Contrast enhancement, used for extending contrasts between classes of Quanternary strata. (b) Operational enhancement such as addition, multiplication, that is values of the same pixel in bands were added or multiplied, quantitated and produced a new image, The result o this method is magnifying the distance of pixel values. For example, values of MSS 5 multiplying the values of MSS 4, it is obvious for distinguishing Q
2 from Q
21 Q
2 fom Q
3, Q
3 from Q
4. (c) Convolution filter, for enhancing the character of edges and texture of classed. After the operation of 3x3 convolution core, the texture of Q
31 And Q
31 are easier to distinguish, Q31 presents point shape and Q32 shows linear shape and the boundaries of classes are more obvious. (d) Ratio enhancement, available method for identifying Quanternary strata, but the key point is how to select optimum ration bands their combination, which are feature parameters of identification and classification, so J-M distance was introduced as quantitation standards for degree of separation for classes, and selections of features and optimum band combinations. For example, the degree of separation between two types of remote sensing data can be represented by J-M distance between two types.
The average J-M distance of different band ratio combinations can be calculated. The bigger J-M distance, the easier separation of classes in Table 2, we listed average J-M distance for different bands combinations.
The definition of J-M distance is as follows:
Where:
J
ij: J-M distance between class I and class j in afeature parameter;.
m
i: mean value of class j;
S
i: Covariance matrix of class j.
S
j: Covariance matrix of class j.
The definition of average J-M distance is as follows:
Where:
Java: average distance;
P(
wi): prior probability of class I;
P(
wj): Prior probability of class j.
Table 2 Average J-M distance of different band combination
| Band combinations |
average J-M distance |
| MSS7, MSS5, MSS4 |
0.4084 |
| MSS7/MSS4/MSS4/MSS5, MSS5/MSS6 |
0.6548 |
| MSS6/MSS7,MSS7/MSS5,MSS5/MSS4 |
0.2503 |
| MSS7, MSS5 |
0.7236 |
| MSS6/MSS5,MSS7/MSS4,MSS6/NSS5 |
0.2765 |
| MSS7+MSS5/MSS7,MSS7+MSS5/MSS5,MSS7+MSS5/MSS |
4 0.2627 |
| K1 K2 K3 |
0.4537 |
(NOTE: G stands for topographic data image, k
1, k
2, k
3 separatedly for three major component images.)
It is known from the Table 2, that the optimal band combination is MSS7, MSS5 and G. The second is MSS7/MSS4, MSS4/MSS5 and MSS5/MSS4. For exaple Q
4, Q
32, Q
31, Q
31, Q
22 and Q
21 repressent different colors in the composed image of MSS7/MSS4, MSS4/MSS5, MSS5/MSS4.
