Neural network code description for general image recognition

The Network format is defined by reading a file. The file format is as follows:

Input Image length input image width hidden layer neuron count output neuron count
Number of different network structures
[Number of hidden layer neurons connected at different locations]
[Position table of input neurons connected by hidden layer neurons]

The following is an example:

24 28 52 1
3
16 32
1 1 1 1 1 1 2 2 2 2 2 3 3 3 3 3 4 4 4 4 4
1 1 1 1 1 1 2 2 2 2 2 3 3 3 3 3 4 4 4 4 4
1 1 1 1 1 1 2 2 2 2 2 3 3 3 3 3 4 4 4 4 4
1 1 1 1 1 1 2 2 2 2 2 3 3 3 3 3 4 4 4 4 4
1 1 1 1 1 1 2 2 2 2 2 3 3 3 3 3 4 4 4 4 4
1 1 1 1 1 1 2 2 2 2 2 3 3 3 3 3 4 4 4 4 4
1 1 1 1 1 1 2 2 2 2 2 3 3 3 3 3 4 4 4 4 4
5 5 5 5 5 6 6 6 6 6 6 6 7 7 7 7 7 7 8 8 8 8
5 5 5 5 5 6 6 6 6 6 6 6 7 7 7 7 7 7 8 8 8 8
5 5 5 5 5 6 6 6 6 6 6 6 7 7 7 7 7 7 8 8 8 8
5 5 5 5 5 6 6 6 6 6 6 6 7 7 7 7 7 7 8 8 8 8
5 5 5 5 5 6 6 6 6 6 6 6 7 7 7 7 7 7 8 8 8 8
5 5 5 5 5 6 6 6 6 6 6 6 7 7 7 7 7 7 8 8 8 8
5 5 5 5 5 6 6 6 6 6 6 6 7 7 7 7 7 7 8 8 8 8
9 9 9 9 9 9 9 10 10 10 10 10 11 11 11 11 12 12 12 12
9 9 9 9 9 9 9 10 10 10 10 10 11 11 11 11 12 12 12 12
9 9 9 9 9 9 9 10 10 10 10 10 11 11 11 11 12 12 12 12
9 9 9 9 9 9 9 10 10 10 10 10 11 11 11 11 12 12 12 12
9 9 9 9 9 9 9 10 10 10 10 10 11 11 11 11 12 12 12 12
9 9 9 9 9 9 9 10 10 10 10 10 11 11 11 11 12 12 12 12
9 9 9 9 9 9 9 10 10 10 10 10 11 11 11 11 12 12 12 12
13 13 13 13 13 13 14 14 14 14 14 15 15 15 15 15 16 16 16 16 16
13 13 13 13 13 13 14 14 14 14 14 15 15 15 15 15 16 16 16 16 16
13 13 13 13 13 13 14 14 14 14 14 15 15 15 15 15 16 16 16 16 16
13 13 13 13 13 13 14 14 14 14 14 15 15 15 15 15 16 16 16 16 16
13 13 13 13 13 13 14 14 14 14 14 15 15 15 15 15 16 16 16 16 16
13 13 13 13 13 13 14 14 14 14 14 15 15 15 15 15 16 16 16 16 16
13 13 13 13 13 13 14 14 14 14 14 15 15 15 15 15 16 16 16 16 16
4 8
1 1 1 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2
1 1 1 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2
1 1 1 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2
1 1 1 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2
1 1 1 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2
1 1 1 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2
1 1 1 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2
1 1 1 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2
1 1 1 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2
1 1 1 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2
1 1 1 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2
1 1 1 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2
1 1 1 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2
1 1 1 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2
3 3 3 3 3 3 3 3 3 3 3 3 3 4 4 4 4 4 4 4 4 4
3 3 3 3 3 3 3 3 3 3 3 3 3 4 4 4 4 4 4 4 4 4
3 3 3 3 3 3 3 3 3 3 3 3 3 4 4 4 4 4 4 4 4 4
3 3 3 3 3 3 3 3 3 3 3 3 3 4 4 4 4 4 4 4 4 4
3 3 3 3 3 3 3 3 3 3 3 3 3 4 4 4 4 4 4 4 4 4
3 3 3 3 3 3 3 3 3 3 3 3 3 4 4 4 4 4 4 4 4 4
3 3 3 3 3 3 3 3 3 3 3 3 3 4 4 4 4 4 4 4 4 4
3 3 3 3 3 3 3 3 3 3 3 3 3 4 4 4 4 4 4 4 4 4
3 3 3 3 3 3 3 3 3 3 3 3 3 4 4 4 4 4 4 4 4 4
3 3 3 3 3 3 3 3 3 3 3 3 3 4 4 4 4 4 4 4 4 4
3 3 3 3 3 3 3 3 3 3 3 3 3 4 4 4 4 4 4 4 4 4
3 3 3 3 3 3 3 3 3 3 3 3 3 4 4 4 4 4 4 4 4 4
3 3 3 3 3 3 3 3 3 3 3 3 3 4 4 4 4 4 4 4 4 4
3 3 3 3 3 3 3 3 3 3 3 3 3 4 4 4 4 4 4 4 4 4
6 12
1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2
2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2
2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2
2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2
3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3
3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3
3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3
3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3
4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4
4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4
4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4
4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4
5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5
5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5
5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5
5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5
6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6
6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6
6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6
6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6
6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6
6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6
6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6
6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6

The following is the program code:

Type

Tsingleextendedarray = array of extended;
Tdoubleextendedarray = array of extended;

Tsamples = packed record
INS: tsingleextendedarray;
Outs: tsingleextendedarray;
End;

Type

Tgraphicbpnn = Class
Private
Procedure backpropagate (T: tsingleextendedarray; n, m: Extended );
Function Update (inputs: tsingleextendedarray): extended;
Public
Samplecounts, testcounts: longint;
Procedure addtotrain (INS, outs: tsingleextendedarray );
Procedure addtotest (INS, outs: tsingleextendedarray );
Procedure savetofile (filename: string );
Procedure loadfromfile (filename: string );
Procedure train (n, m: Extended );
Function Init (filename: string): Boolean;
Function predict (INS: tsingleextendedarray): extended;
Function Test: extended;
Destructor destroy; override;
Private
Ni, NH, No: longint;
AI, ah, AO, output_deltas, hidden_deltas: tsingleextendedarray;
WI, wo, CI, CO: tdoubleextendedarray;
Connections: array of Boolean;
Samples: array of tsamples;
Testset: array of tsamples;
End;

Implementation

Function tgraphicbpnn. INIT (filename: string): Boolean;
VaR
I, J, K, Fi, FJ: longint;
NIW, institutes, ropmax, ropnum, roptypes: longint;
Ropmap: array of longint;
Begin
Assignfile (input, filename );
Reset (input );
Readln (input, NIW, institutes, NH, no );
Ni: = NIW * NI;
Setlength (AI, Ni );
Setlength (ah, NH );
Setlength (Ao, no );
For I: = 0 to ni-1 do Ai [I]: = 1;
For I: = 0 to NH-1 do Ah [I]: = 1;
For I: = 0 to no-1 do AO [I]: = 1;

Setlength (WI, Ni, NH );
Setlength (WO, NH, no );
Setlength (CI, Ni, NH );
Setlength (CO, NH, no );
Setlength (connections, Ni, NH );

For I: = 0 to ni-1 do
For J: = 0 to NH-1 do
Connections [I, j]: = false;

Readln (roptypes); FJ: = 0;
For K: = 1 to roptypes do begin
Readln (ropmax, ropnum );
Setlength (ropmap, Ni );
FI: = 0;
For I: = 1 to institutes do begin
For J: = 1 to NIW do begin
Read (ropmap [fi]);
INC (FI );
End;
Readln;
End;
FI: = 0;
For I: = 1 to ropnum do begin
INC (FI );
If fi> ropmax then FI: = 1;
For J: = 0 to ni-1 do
If ropmap [J] = Fi then connections [J, FJ]: = true;
INC (fj );
End;
End;

Setlength (output_deltas, no );
Setlength (hidden_deltas, NH );

Randomize;
For I: = 0 to ni-1 do
For J: = 0 to NH-1 do begin
CI [I, j]: = 0;
Wi [I, j]: = random (40000)/10000-2;
End;

For I: = 0 to NH-1 do
For J: = 0 to no-1 do begin
CO [I, j]: = 0;
WO [I, j]: = random (40000)/10000-2;
End;

Setlength (samples, $100); setlength (testset, $100 );
Samplecounts: = 0; testcounts: = 0;
Closefile (input );
End;

Procedure tgraphicbpnn. backpropagate (T: tsingleextendedarray; n, m: Extended );
VaR
I, J, K: longint;
Sum, change: extended;
Begin
For I: = 0 to no-1 do
Output_deltas [I]: = AO [I] * (1-ao [I]) * (T [I]-ao [I]);

For J: = 0 to NH-1 do begin
Sum: = 0;
For K: = 0 to no-1 do
Sum: = sum + output_deltas [k] * WO [j, k];
Hidden_deltas [J]: = Ah [J] * (1-Ah [J]) * sum;
End;

For J: = 0 to NH-1 do
For K: = 0 to no-1 do begin
Change: = output_deltas [k] * Ah [J];
WO [j, k]: = Wo [j, k] + N * change + M * CO [j, k];
CO [j, k]: = change;
End;

For I: = 0 to ni-1 do
For J: = 0 to NH-1 do
If connections [I, j] Then begin
Change: = hidden_deltas [J] * Ai [I];
Wi [I, j]: = wi [I, j] + N * change + M * CI [I, j];
CI [I, j]: = change;
End;

End;

Function tgraphicbpnn. Update (inputs: tsingleextendedarray): extended;
VaR
I, J, K: longint;
Sum: extended;
Begin
For I: = 0 to ni-1 do
Ai [I]: = inputs [I];
For J: = 0 to NH-1 do begin
Sum: = 0;
For I: = 0 to ni-1 do
If connections [I, j] Then
Sum: = sum + AI [I] * WI [I, j];
Ah [J]: = 1/(1 + exp (-sum ));
End;
For K: = 0 to no-1 do begin
Sum: = 0;
For J: = 0 to NH-1 do
Sum: = sum + Ah [J] * WO [j, k];
AO [k]: = 1/(1 + exp (-sum ));
End;
Update: = AO [0];
End;

Procedure tgraphicbpnn. Train (n, m: Extended );
VaR I: longint;
Begin
For I: = 0 to samplecounts-1 do begin
Update (samples [I]. INS );
Backpropagate (samples [I]. outs, n, m );
End;
End;

Procedure tgraphicbpnn. addtotrain (INS, outs: tsingleextendedarray );
VaR I: longint;
Begin
If samplecounts> high (samples) Then setlength (samples, samplecounts ++ $100 );
Setlength (samples [samplecounts]. INS, Ni );
Setlength (samples [samplecounts]. outs, no );
For I: = 0 to ni-1 do samples [samplecounts]. INS [I]: = INS [I];
For I: = 0 to no-1 do samples [samplecounts]. outs [I]: = outs [I];
INC (samplecounts );
End;

Procedure tgraphicbpnn. addtotest (INS, outs: tsingleextendedarray );
VaR I: longint;
Begin
If testcounts> high (testset) Then setlength (testset, testcounts + $100 );
Setlength (testset [testcounts]. INS, Ni );
Setlength (testset [testcounts]. outs, no );
For I: = 0 to ni-1 do testset [testcounts]. INS [I]: = INS [I];
For I: = 0 to no-1 do testset [testcounts]. outs [I]: = outs [I];
INC (testcounts );
End;

Procedure tgraphicbpnn. savetofile (filename: string );
VaR
I, J, K: longint;
Savestream: tmemorystream;
Begin
Savestream: = tmemorystream. Create;
Savestream. Seek (0, 0 );
For I: = 0 to ni-1 do
For J: = 0 to NH-1 do begin
Savestream. Write (WI [I, j], sizeof (WI [I, j]);
Savestream. Write (CI [I, j], sizeof (CI [I, j]);
End;
For J: = 0 to NH-1 do
For K: = 0 to no-1 do begin
Savestream. Write (WO [J, K], sizeof (WO [j, k]);
Savestream. Write (CO [J, K], sizeof (CO [j, k]);
End;
Savestream. savetofile (filename );
Savestream. Free;
End;

Procedure tgraphicbpnn. loadfromfile (filename: string );
VaR
I, J, K: longint;
Readstream: tmemorystream;
Begin
Readstream: = tmemorystream. Create;
Readstream. loadfromfile (filename );
Readstream. Seek (0, 0 );
For I: = 0 to ni-1 do
For J: = 0 to NH-1 do begin
Readstream. Read (WI [I, j], sizeof (WI [I, j]);
Readstream. Read (CI [I, j], sizeof (CI [I, j]);
End;
For J: = 0 to NH-1 do
For K: = 0 to no-1 do begin
Readstream. Read (WO [J, K], sizeof (WO [j, k]);
Readstream. Read (CO [J, K], sizeof (CO [j, k]);
End;
Readstream. Free;
End;

Function tgraphicbpnn. predict (INS: tsingleextendedarray): extended;
Begin
Try
Predict: = Update (INS );
Except
Predict: = 0;
End;
End;

Function tgraphicbpnn. Test: extended;
VaR
Preret: extended;
I, counts, RET: longint;
Begin
Counts: = 0;
For I: = 0 to testcounts-1 do begin
Preret: = predict (testset [I]. INS );
If preret> 0.5 then RET: = 1 else RET: = 0;
If ret = testset [I]. outs [0] Then Inc (counts );
End;
Result: = counts/testcounts;
End;

Destructor tgraphicbpnn. Destroy;
Begin
Setlength (AI, 0 );
Setlength (ah, 0 );
Setlength (Ao, 0 );
Setlength (output_deltas, 0 );
Setlength (hidden_deltas, 0 );
Setlength (WI, 0, 0 );
Setlength (WO, 0, 0 );
Setlength (CI, 0, 0 );
Setlength (CO, 0, 0 );
Setlength (connections, 0, 0 );
Setlength (samples, 0 );
Inherited;
End;